2011-05-03T15:16:32 *** rtemsLogger has joined #rtems 2011-05-03T15:36:32 *** peerst has joined #rtems 2011-05-03T16:25:17 *** DrJoel has joined #rtems 2011-05-03T16:25:17 *** ChanServ sets mode: +o DrJoel 2011-05-03T16:26:48 *** DrJoel changes topic to "OAR has power again. Running checks on rtems.org but no problems so far." 2011-05-03T16:27:41 *** DrJoel changes topic to "OAR has power again. Running checks on rtems.org but no problems so far. http://www.rtemsramblings.com for more news on storm" 2011-05-03T16:29:18 *** DrJoel changes topic to "OAR has power again. Running checks on rtems.org but no problems so far. mail.oarcorp.com is still down as best I can tell. Dawn to dusk curfew in effect so we are doing what we can remotely until the morning when we can go back onsite. http://www.rtemsramblings.com for more news on storm" 2011-05-03T16:50:22 *** DrJoel has quit IRC 2011-05-03T17:02:44 rtemsramblings.com? 2011-05-03T17:31:31 *** projectgus has joined #rtems 2011-05-03T17:33:04 that domain is not even registered am i missing something? 2011-05-03T17:33:41 This works .... http://rtemsramblings.blogspot.com/ 2011-05-03T17:37:52 *** DrJoel has joined #rtems 2011-05-03T17:37:52 *** ChanServ sets mode: +o DrJoel 2011-05-03T17:38:03 Hi Joel 2011-05-03T17:38:28 hey kiwichris .. once things settle out, we have some post game analysis to do. :-D 2011-05-03T17:38:39 I want a more complete mirror site 2011-05-03T17:38:40 Yeah I agree. 2011-05-03T17:38:44 Sure. 2011-05-03T17:39:00 Is www.rtemsramblings.com real ? 2011-05-03T17:39:01 I never would have considered a 6 day outage in one of the U.S.'s largest research park a remote possibility 2011-05-03T17:39:28 Grrr... 2011-05-03T17:39:48 *** DrJoel changes topic to "OAR has power again. Running checks on rtems.org but no problems so far. mail.oarcorp.com is still down as best I can tell. Dawn to dusk curfew in effect so we are doing what we can remotely until the morning when we can go back onsite. http://rtemsramblings.blogspot.com for more news on the storm" 2011-05-03T17:40:12 It is kind of difficult to imagine the scale of the disaster. It is not something you typically plan for. 2011-05-03T17:40:38 no. normally rtems.info being a complete mirror would be sufficient for short term use 2011-05-03T17:40:52 and if more bandwidth needed, the machine gets moved or more bandwidth bought 2011-05-03T17:41:20 We need to get feedback on the yum repo mirror structure from Ralf. 2011-05-03T17:41:52 and ... mysql.. mail.. etc.. loss of updates from lab machines would be OK temporarily 2011-05-03T17:42:46 Mail is a it of an issue because I use the rtems.org address a lot. It we have a mirror could it be a secondary MX record 2011-05-03T17:42:58 those are the issues that we have to nail.. 2011-05-03T17:43:23 also we have never went through the motions of temporarily changing the DNS with ISP. 2011-05-03T17:43:28 all of that needs to be nailed down. 2011-05-03T17:43:37 Do you think an Amazon EC2 cloud might be OK? 2011-05-03T17:44:53 No idea. I have not looked at this service before. 2011-05-03T17:45:55 me either.. would like to buy new 2U, clean up current machine as a nice 1U with new disks, sync it, make sure it mirrors locally. Then ship it somewhere. Hopefully to an open source friendly place that is low cost. OSU might be an option 2011-05-03T17:46:29 NIce idea. 2011-05-03T17:46:57 It may also be possible to do a trade with another FOSS project .. we host a 1U machine for us.. they host a 1U machine for them. 2011-05-03T17:47:10 We keep upgrading rtems.org all the time. How would this be handled ? 2011-05-03T17:47:45 if we can remotely access it, then via ssh. Some hands-on admin might be needed. We would need a local for that. All details to worry about. 2011-05-03T17:48:13 Can a Fedora upgrade be done via ssh ? 2011-05-03T17:48:16 Easiest thing might be to make rtems.info a more full mirror.. 2011-05-03T17:48:33 Maybe the root disk is on a SATA plug bay. 2011-05-03T17:48:45 yes.. if all goes well.. you can upgrade via net. Jeff does it on machines with only CDs 2011-05-03T17:49:36 Since I am already mirroring content and have full control over rtems.info, I also can do upgrades at the same time. It is probably the easiest machine to be a warm failover 2011-05-03T17:49:45 Has rtems.org been upgraded by just ssh recently ? 2011-05-03T17:49:51 We would need to find an ISP to move it to for failure 2011-05-03T17:50:01 I ran yum upgrade earlier today 2011-05-03T17:50:13 Did your ISP go down as well ? 2011-05-03T17:50:52 I am not so concerned about updates rather upgrades to a new version of Fedora 2011-05-03T17:50:54 No idea really because I had no power. They could have been up.. wouldn't have helped me. 2011-05-03T17:51:22 I understand.. we also need practice at the mirroring. 2011-05-03T18:03:02 i can help setup mirrors 2011-05-03T18:03:12 we can setup alternate MTA tanks to ensure delivery 2011-05-03T18:03:30 DrJoel: i'll get you an email or two for osu perhaps the 1u can be sent there 2011-05-03T18:03:58 Thanks verm__ 2011-05-03T18:04:18 FWIW my ISP can provide me with 50Mbps down. Not sure about the uplink 2011-05-03T18:05:50 Business Ultimate 2011-05-03T18:05:50 50 Mbps down/5 Mbps up 2011-05-03T18:05:50 Business Max 30 Mbps down/5 Mbps up 2011-05-03T18:05:50 Business Select 16 Mbps down/2 Mbps up 2011-05-03T18:05:50 Business Basic 8 Mbps down/1 Mbps up 2011-05-03T18:06:02 I have basic and it is usually 1.5 up 2011-05-03T18:06:06 * DrJoel is walking away 2011-05-03T18:21:20 ah 2011-05-03T18:53:53 *** DrJoel has quit IRC 2011-05-03T19:14:47 *** xiangfu has joined #rtems 2011-05-03T19:54:54 kiwichris: offhand do you know of any spare machines at OAR? 2011-05-03T19:55:14 For a build bot ? 2011-05-03T19:55:22 yeah 2011-05-03T19:55:27 rtbf32a 2011-05-03T19:55:31 doesn't need to be paricularly high powered 2011-05-03T19:55:40 This is what I currently use. 2011-05-03T19:55:57 PLently of disk under /home2 I think 2011-05-03T19:56:09 is that the same machine the builds run on? what else does it do? 2011-05-03T19:56:40 It does the CVS email patch to mailing list. That is all. 2011-05-03T19:56:44 I know of 2011-05-03T19:56:50 oh that's perfect then 2011-05-03T19:56:55 i'll put the master on that machine 2011-05-03T19:56:57 Yeah. 2011-05-03T19:57:00 Great 2011-05-03T19:57:12 If you need any installed let me know. 2011-05-03T19:57:15 rtbf32a appears to be down at the moment 2011-05-03T19:58:58 kiwichris: what OS is it? 2011-05-03T19:59:10 Fedora something :) 2011-05-03T19:59:16 It is a 32bit machine. 2011-05-03T19:59:42 It is a 64bit CPU but we have a 32bit machine about to allow specific testing. 2011-05-03T19:59:49 that's fine i'm sure it'll be enough 2011-05-03T20:00:10 It is a fast machine. 2011-05-03T20:00:15 great 2011-05-03T20:01:48 whoa fedora sure has an old version of buildbot 2011-05-03T20:02:24 No idea. Maybe Ralf knows. 2011-05-03T20:02:35 i checked the online packagedb 2011-05-03T20:02:48 Is it FC14 ? 2011-05-03T20:02:59 yeah, 13,14,15,devel 2011-05-03T20:03:01 https://admin.fedoraproject.org/pkgdb/acls/name/buildbot 2011-05-03T20:05:28 ok there's no way i can use anything as far as fedora packages go it's all too old 2011-05-03T20:05:43 i'll use python virtualenv and run it that way it'll be much easier 2011-05-03T20:06:02 but i'd like at least python 2.7, mysql, mysqldb installed 2011-05-03T20:06:08 Ok 2011-05-03T20:06:12 Hang on 2011-05-03T20:06:45 is the machine currently up? 2011-05-03T20:07:10 Does not look like it. 2011-05-03T20:07:33 I think the lab maybe down. Joel said in his blog something smelt burnt 2011-05-03T20:07:55 whoa 2011-05-03T20:08:11 No ping response. :( 2011-05-03T20:08:18 yeah i saw no arp at all 2011-05-03T20:08:33 It will need to wait until tomorrow when Joel heads in. 2011-05-03T20:08:33 we'll have to wait until tomorrow due to the curfew 2011-05-03T20:08:39 Yeah 2011-05-03T20:11:34 *** peerst has quit IRC 2011-05-03T20:41:56 kiwichris - so, unexpectedly, it looks like my bug has something to do with POSIX signal handlers in RTEMS 2011-05-03T20:42:21 Readline installs half-a-dozen wrapper signal handlers while it's waiting on stdin, and something in that process seems to trip off my bug 2011-05-03T20:42:35 it doesn't seem to be any particular signal handler though, something in the entire process is breaking it somehow 2011-05-03T20:42:54 for now I'm just building Readline with HAVE_SIGNALS turned off, because I don't have much more time to spend on it 2011-05-03T20:43:09 but I thought I'd mention it just in case it rings any bells, I'm on RTEMS 4.10, or in case it might be helpful info to share around 2011-05-03T21:08:21 None with me. I do not use signals with RTEMS nor readline. It may pay to post this to the list. It is back now. 2011-05-03T21:28:51 thanks kiwichris. I posted it to the EPICS list and I got a lead about EPICS base possibly abusing SIGALRM signal handlers 2011-05-03T21:28:54 at one point 2011-05-03T21:29:13 ok 2011-05-03T21:29:22 although (a) that may be fixed, and (b) I still had the issue even when I took out the SIGALRM remapping in readline, so it's still a bit up in the air 2011-05-03T21:29:34 I'll post to RTEMS in a day or two if nothing comes out of the EPICS-side investigations 2011-05-03T21:30:04 Sure. If you feel it is RTEMS please raise a PR. I know Joel wants to know about any POSIX related issues. 2011-05-03T21:30:21 OK, will do 2011-05-03T21:58:01 kiwichris: is there any area outside of the lab where i can look at the current test framework? 2011-05-03T21:58:57 The wiki I think has something. 2011-05-03T21:59:41 hmm i meant more from the scripts that we use to run 2011-05-03T21:59:45 use to run it that is 2011-05-03T21:59:58 Ah sorry. Yes CVS under rtems-testing 2011-05-03T22:00:13 I have a patch to turn this into an autoconf/automake package 2011-05-03T22:00:37 It then can provide 'rtems-test' as a script that handles all the mess for running simulators. 2011-05-03T22:00:44 nice 2011-05-03T22:01:15 I needed to make some changes which Joel asked me to do but I have not done this. I have the email. 2011-05-03T22:01:38 Currently rtems-testing is a large number of scripts which is not easy to handle. 2011-05-03T22:01:48 do we actually have a physical copy of coverity? 2011-05-03T22:01:54 No 2011-05-03T22:01:59 bummer 2011-05-03T22:02:12 We are given access, or Joel is because we are open source. 2011-05-03T22:02:21 yeah through their test system i know 2011-05-03T22:02:32 Yeah 2011-05-03T22:02:34 i saw a coverity dir here and got curious :) 2011-05-03T22:03:20 so, basically what we'll want to do is build rtems for a platform 2011-05-03T22:03:25 then run all the tests in order for that platform 2011-05-03T22:03:26 ? 2011-05-03T22:03:30 Yes 2011-05-03T22:03:41 are there any situations where the tests are run without rtems being re-built? 2011-05-03T22:03:50 or more to the point, are there any cases that we care about where this might happen 2011-05-03T22:04:17 Maybe in time yes :) 2011-05-03T22:04:24 heh ok 2011-05-03T22:04:32 I am currently adding dynamic loading which will allow this 2011-05-03T22:05:07 The rub is the ability for the simulator to allow this. 2011-05-03T22:05:40 For example qemu allows disk access via IDE for i386. Here a single RTEMS base build could run various tests 2011-05-03T22:06:00 How-ever the SIS gdb simulator does not have this. 2011-05-03T22:08:39 hmm that sucks 2011-05-03T22:08:48 as i recall rtems is entirely cross-built correct? 2011-05-03T22:09:05 Not really because it is what we face with real hardware and real users. 2011-05-03T22:09:12 Yes 2011-05-03T22:09:30 if i'm seeing this correctly, the entire toolchain is built first, then rtems is built then it is tested? 2011-05-03T22:09:38 Yes. 2011-05-03T22:09:47 The tool chain is all handled via yum 2011-05-03T22:09:56 We do not need to worry about this. 2011-05-03T22:09:59 the testing, are these out-of-band tests or does it require the rtems instance to be launched first then the tests run inside? 2011-05-03T22:10:14 Not at the moment. 2011-05-03T22:10:21 kiwichris: that's good to know, i hate toolchain nonsense it gets annoying fast :) 2011-05-03T22:10:39 kiwichris: not to which, they're out of band or an instance is required? 2011-05-03T22:10:43 The model you raised is the dynamic loader one. 2011-05-03T22:10:59 well it's also possible for me to install rtems 2011-05-03T22:11:02 run it, set it up 2011-05-03T22:11:03 We create static executables with RTEMS + test 2011-05-03T22:11:07 copy the tests inside and run the tests remotely 2011-05-03T22:11:26 i've done this many times before when testing OS and software for embedded systems 2011-05-03T22:11:28 With the build bot ? 2011-05-03T22:11:45 well i'd write something outside of it, setting up buildbot to trigger these events is trivial 2011-05-03T22:12:00 it would be seperate steps, first step would be to install + config the rtems installation 2011-05-03T22:12:10 second step to copy all the data over, next one to run the tests, last step would be cleanup 2011-05-03T22:12:23 Ok. All that is needed is the simulators and the various BSP specific scripts (held in rtems-testing) 2011-05-03T22:12:26 we can install to a VM, or to a physical machine -- physical machines are probably better as this is closer to real-world any kind of hardware is fine 2011-05-03T22:12:52 Not sure I follow. 2011-05-03T22:13:38 after rtems is built, do you actually install it somewhere in order for tests to run inside of it? 2011-05-03T22:14:11 No you can do this if you can locate the executables in the build tree. 2011-05-03T22:14:33 Currently they are *.exe but this needs to change and this is where things get complicated. 2011-05-03T22:14:43 exe? what? 2011-05-03T22:14:51 I would not install RTEMS after the build. 2011-05-03T22:14:58 what are exe? 2011-05-03T22:15:24 Current an RTEMS configured to build tests creates executables, eg hello.exe which is an ELF file with RTEMS and test code 2011-05-03T22:15:43 oh i see how you're doing it, ok 2011-05-03T22:15:57 The .exe was added and it is a mistake because it effect automake 2011-05-03T22:16:15 ah, i see 2011-05-03T22:16:27 So I want to remove the .exe from the executable file names. 2011-05-03T22:16:36 yeah that would make sense :) 2011-05-03T22:16:40 how many target platforms are there? 2011-05-03T22:16:54 This how-ever means given a build tree you have no idea which files are the actual executables. 2011-05-03T22:17:37 http://www.rtems.org/ftp/pub/rtems/people/chrisj/pq/html/ 2011-05-03T22:17:39 what do you mean? any executables should have +x on them 2011-05-03T22:17:51 targets 14, BSP 122 2011-05-03T22:18:17 Well should they ? They do not run on the host, or any host 2011-05-03T22:18:43 hmm 2011-05-03T22:18:46 In this case what does a +x bit mean ? 2011-05-03T22:19:01 i see why you added *.exe now 2011-05-03T22:19:07 +x is the executable bit in unix 2011-05-03T22:19:38 wow so we're going to need 122 builders that's a lot more than i was expecting, nice 2011-05-03T22:19:41 The problem is automake uses the linker default output to set $(EXEEXT) and our default is a.out so automake gets it wrong 2011-05-03T22:20:06 Also .exe is a filename that is looked for on Windows to execute. Check cygwin and MSYS code. 2011-05-03T22:20:36 A solution is to use autotest and get make to do the testing for us. 2011-05-03T22:20:47 This is where I have got to and stopped. 2011-05-03T22:21:30 can you tell me what is involved in testing? 2011-05-03T22:21:34 is it a simple pass/fail system or? 2011-05-03T22:21:41 Yes I know +x is this on Unix but a SIS simulator does not have a file system. :) 2011-05-03T22:21:59 yeah i know i realised what you meant after saying they're not executed locally :) 2011-05-03T22:22:08 this is all becoming a lot more clear for me thank you 2011-05-03T22:22:13 Currently run the test and if it reaches the end statement it is ok. This is a poor pass evaluation. 2011-05-03T22:22:54 My changes to rtems-testing to automake etc and then rtems-test script is to allow the ability to handle the results better 2011-05-03T22:24:11 any particular reason you chose automake versus a scripting language? 2011-05-03T22:25:06 Ah ok. Need to back up a bit. Autoconf and automake is only to configure and build rtems-testing on the host. It does not relate to what is contains. 2011-05-03T22:25:31 At the moment rtems-test is a shell script. It could be a Python or perl script. 2011-05-03T22:25:53 right, i know that part -- what you're saying is the executor of the tests is now automake? 2011-05-03T22:26:39 No. Back one more level. It is to build the rtems-testing package so you can install it and then use it. 2011-05-03T22:27:03 Once you have rtems-testing installed and the tools installed you can built RTEMS. 2011-05-03T22:27:28 And then you can use rtems-test to run a specific executable on simulated hardware. 2011-05-03T22:27:54 Oh and all 122 BSPs will not be able to be tested. A number relate to specific hardware 2011-05-03T22:28:29 what is a 'BSP' btw? 2011-05-03T22:28:31 The 2nd part is how to run a specific set of RTEMS built tests. 2011-05-03T22:28:37 Board Support Package 2011-05-03T22:28:52 Old term from the before the 'net 2011-05-03T22:28:54 oh, of course 2011-05-03T22:29:05 yes i know, BSP means so many things now 2011-05-03T22:29:17 Like everything. 2011-05-03T22:29:22 hehe yeah 2011-05-03T22:29:33 so what is used to run the individual tests right now? 2011-05-03T22:29:53 So we need a rtems-testing sorted. Then we need to think about RTEMS itself. It is here that autotest as a solution was raised. 2011-05-03T22:30:10 right now i'm following what's going on 2011-05-03T22:30:28 why is autotest more useful than say, a python script? 2011-05-03T22:31:10 Not sure. I have not look. One advantage is the knowledge of what is an executable because it is integrated to the build system. 2011-05-03T22:31:17 And this relates back to the .exe issue. 2011-05-03T22:31:41 right but we can fix that by using automake to generate a manifest file 2011-05-03T22:32:03 Yes. Then there is another complication :) Don;t you love them. 2011-05-03T22:32:11 Some tests can run and some cannot. 2011-05-03T22:32:15 heh nothing is ever easy :) 2011-05-03T22:32:16 For each BSP. 2011-05-03T22:32:28 how is this tracked currently? 2011-05-03T22:32:31 For example on some hardware file system tests use too much memory 2011-05-03T22:32:48 With a horrible hack I added as a work around and has stayed. 2011-05-03T22:33:08 oh, nice. 2011-05-03T22:33:25 is manually tracking what can run and not run an option? 2011-05-03T22:33:37 We think we need something in the BSP itself that lists tests that are out. 2011-05-03T22:33:54 No. It needs to be in the BSP. The BSP developer will know. 2011-05-03T22:34:08 So yes manually as a file in the BSP. 2011-05-03T22:34:29 what i mean is if test X cannot run in the BSP then is manually marking the test is "cannot run on BSP XX due to YY" an OK solution? 2011-05-03T22:34:39 Yes. 2011-05-03T22:34:56 how many tests are there roughly? 2011-05-03T22:35:12 Hundreds 2011-05-03T22:35:19 Look in teststuite 2011-05-03T22:35:25 from the top level. 2011-05-03T22:35:26 so we need a web application for this? 2011-05-03T22:35:49 where is testsuite? 2011-05-03T22:35:52 I am not sure. 2011-05-03T22:36:09 It would be nice to know if the fails changed by a threshold. 2011-05-03T22:36:22 In the RTEMS source tree 2011-05-03T22:36:35 great, exactly where i didn't think it would be. 2011-05-03T22:36:50 :) 2011-05-03T22:36:58 ok yeah this is a lot of tests... 2011-05-03T22:37:19 It is growing all the time. 2011-05-03T22:37:27 each one of these tests creates a single executable right? 2011-05-03T22:37:33 Yes 2011-05-03T22:37:42 RTEMS is a static operating system (at the moment) 2011-05-03T22:38:03 alright, so my proposed solution 2011-05-03T22:38:21 hmm actually, first of all 2011-05-03T22:38:29 what's that evil hack you were talking about a moment ago? :) 2011-05-03T22:39:58 *** arvind_khadri has joined #rtems 2011-05-03T22:40:38 Every BSP has a bsp.h file. I added a BSP_SMALL_MEMORY (or something like this) define to the ones that fails than defined out the whole test if present. 2011-05-03T22:40:55 Hmmm, how I have publicly documented I did it :) 2011-05-03T22:41:32 why did you choose to do this in source 2011-05-03T22:41:40 versus simply not running the test on that bsp? 2011-05-03T22:42:09 Because it was too difficult to do in the build system at the time. Ralf was moving config data to each BSP. 2011-05-03T22:42:30 That is now done so we can look at a better way to handle this 2011-05-03T22:42:53 autotest is a huge test suite for testing, i'm not sure if you need it 2011-05-03T22:43:12 if all we're doing is running binaries on the BSP it seems like a python script is in order 2011-05-03T22:43:44 we can also maintain a list of "disabled" tests on a per bsp+test basis 2011-05-03T22:43:49 It is but can it help with the pass/fail parsing of the output ? 2011-05-03T22:44:04 how much is there to parse? 2011-05-03T22:44:10 surely the output is uniform for the tests? 2011-05-03T22:44:31 Hmmm I am not so sure. 2011-05-03T22:44:33 we could also add a "machine readable" output format for the tests 2011-05-03T22:44:40 which can be easily parsed using python 2011-05-03T22:44:42 Yes 2011-05-03T22:44:44 maybe output JSON or something 2011-05-03T22:44:55 The targets are small. 2011-05-03T22:45:01 that run these tests 2011-05-03T22:45:28 are these hardware or software simulators or a combination? 2011-05-03T22:45:34 I agree with you. I have not seen a specific reason to use autotest. 2011-05-03T22:45:48 i can write a python system to do this very easily 2011-05-03T22:45:59 Currently just software, but that simulate targets with small memory 2011-05-03T22:47:11 I need to still consider autotest until it proves unsuitable. 2011-05-03T22:47:46 If we need autotest I suspect Ralf will have to provide a package to install via yum. 2011-05-03T22:47:55 sure, my point on that is an entire test suite 2011-05-03T22:47:58 which you don't need 2011-05-03T22:48:14 And a good point. 2011-05-03T22:48:28 it's huge too 2011-05-03T22:48:31 i think it's over 100MB now 2011-05-03T22:48:35 35MB compressed 2011-05-03T22:48:40 and comes with everything you need 2011-05-03T22:49:04 Oh I see it contains many many tests which we have in RTEMS. 2011-05-03T22:50:17 we could still use it if we want to use the reporting 2011-05-03T22:50:21 and other features 2011-05-03T22:50:41 If they work then that is a + 2011-05-03T22:50:41 a python script would be handy if we want to have the ability for users to run tests 2011-05-03T22:51:13 But we have to maintain the script. 2011-05-03T22:51:41 autotest is also python and you have to maintain config files and setups which are also python 2011-05-03T22:51:51 My ideal outcome is a way to create a report that states how the BSP went. 2011-05-03T22:52:06 It autotest implemented in Python ? 2011-05-03T22:52:10 yep 2011-05-03T22:52:16 Hee that is interesting. 2011-05-03T22:52:21 it has a WWW frontend and everything 2011-05-03T22:52:47 if we use it what we would do is use buildbot to build rtems 2011-05-03T22:52:55 then launch the autotest results, buildbot will wait for it to complete ok 2011-05-03T22:53:05 I am talking about this http://www.gnu.org/s/hello/manual/autoconf/Using-Autotest.html 2011-05-03T22:53:09 hmmm, i wish you guys had a freebsd box i could use 2011-05-03T22:53:53 oh haha 2011-05-03T22:54:02 http://autotest.kernel.org/ 2011-05-03T22:54:05 this is what autotest is for me. 2011-05-03T22:54:19 That explains the Python bit. 2011-05-03T22:54:25 many large companies use it, 2011-05-03T22:54:28 Sure. 2011-05-03T22:55:04 Yeap that one is way to heavy for our needs. 2011-05-03T22:55:19 well you also get the fancy reporting, trending and other things with it 2011-05-03T22:55:44 I want users to do this with a standard RTEMS build 2011-05-03T22:55:53 And get a report. 2011-05-03T22:55:58 and that's exactly why i thought autotest was a bad idea 2011-05-03T22:56:01 An HTML file or PDF. 2011-05-03T22:56:22 pdf would require special tools to be installed 2011-05-03T22:56:52 we can output a json file then supply a templated .html file that loads it and displays the data using javascript rather than generating html directly 2011-05-03T22:57:00 this would also allow us to copy these json files to the web and do the same thing 2011-05-03T22:57:21 doing it this way means you only have to create the html template once, you'd be able to modify the json output as you want 2011-05-03T22:57:25 From the target or the test script ? 2011-05-03T22:57:36 from the test script, 2011-05-03T22:57:48 Great. Can this create a PDF ? 2011-05-03T22:57:49 first run all the tests and collect the data, output results at the end 2011-05-03T22:58:07 Great 2011-05-03T22:58:19 well, whoever wants to create a PDF will need those types of tools installed 2011-05-03T22:58:38 i could create a pdf but i can't promise to have enough time to do that anytime soon 2011-05-03T22:58:43 i can make a nice html page and text output though 2011-05-03T22:58:49 Fine. 2011-05-03T22:58:57 i'll document the python script very nicely it will be easy to maintain 2011-05-03T22:59:19 Excellent. I know Python so can be a help. 2011-05-03T23:01:31 so, these tests generate output 2011-05-03T23:01:35 which is compared to the *.scn file? 2011-05-03T23:01:57 Yes but some values can vary. 2011-05-03T23:02:15 how are these marked? 2011-05-03T23:02:21 For example the id value for a task may be different on different BSP 2011-05-03T23:02:25 They are not. 2011-05-03T23:02:32 wouldn't that result in a fail then? 2011-05-03T23:02:45 They are not compared for this reason. 2011-05-03T23:03:02 where is the current system that does the comparison? 2011-05-03T23:03:04 I think we need a tagged file 2011-05-03T23:03:27 It does not. It looks for a valid end. The evaluation of pass is weak. 2011-05-03T23:03:55 i see what you mean now, changing the output of the tests would be a very large task 2011-05-03T23:05:05 for now -- before we could consider doing this in a saner way 2011-05-03T23:05:27 what i'll do is have the script open the *.scn file and capture stdout and do a comparison 2011-05-03T23:05:51 i'll load all this data into a database so we can do trending otherwise generate a report at the end 2011-05-03T23:06:12 we can do trending later on it's easy enough to add 2011-05-03T23:06:23 once we have a line-level comparison we can create an exclusion list on a per-test basis 2011-05-03T23:06:24 Great. I also look for a marked up file and use that with a more solid result. This way we can move over. 2011-05-03T23:07:08 I think you have a fair idea of the issues involved. 2011-05-03T23:07:12 this shouldn't be too bads 2011-05-03T23:07:14 -s 2011-05-03T23:07:26 first thing we'll do is get the build under buildbot 2011-05-03T23:07:32 and write the python script to run the tests once that's running well 2011-05-03T23:07:40 I will look at my rtems-testing changes again. 2011-05-03T23:07:43 we'll figure out how to do trending and other fancy reporting measures 2011-05-03T23:09:02 why did you add $(EXEXT) to your target info? 2011-05-03T23:09:07 automake should handle this automatically 2011-05-03T23:10:00 Automake sets $(EXEEXT) using the results of the linker default output file name. Ours in a.out not a.exe. 2011-05-03T23:10:34 you should be able to override that though 2011-05-03T23:13:01 kiwichris: does rtems build on windows? do you ever need exeext support? 2011-05-03T23:13:13 Yes. 2011-05-03T23:13:21 for what platform(s)? 2011-05-03T23:13:26 MinGW 2011-05-03T23:14:14 The $(EXEEXT) issue is complicated. 2011-05-03T23:14:42 Basically using it requires you add all your own rules to build and this defaults the purpose of using automake 2011-05-03T23:15:13 you should be able to override how automake finds the exension though 2011-05-03T23:16:04 I could not which does not mean you cannot do this. 2011-05-03T23:16:28 i'm fairly certain you can, i should try building rtems on freebs dso i can get an environment going 2011-05-03T23:17:31 have a look at this ..... 2011-05-03T23:18:01 http://wiki.rtems.org/wiki/index.php/SpecBuilder 2011-05-03T23:18:30 It can read the spec files in RTEMS and build the tools. 2011-05-03T23:18:39 i can't belive i'm checking out a CVS repo it's been years 2011-05-03T23:18:53 I still walk to work ;) 2011-05-03T23:19:03 i don't have a license and have never driven a car 2011-05-03T23:19:15 Nice 2011-05-03T23:19:27 if i ever get married i'm sure this will change 2011-05-03T23:19:40 but i'll stick to my hard-headed philosphy that not everyone in this world needs a car :) 2011-05-03T23:19:58 Most do not. 2011-05-03T23:20:12 that is not how most feel though 2011-05-03T23:20:42 * kiwichris walks away 2011-05-03T23:23:29 *** arvind_khadri has quit IRC 2011-05-03T23:32:35 verm__, if you have any SpecBuilder issues let me know 2011-05-03T23:33:14 hmmm 2011-05-03T23:33:32 * verm__ tries 2011-05-03T23:34:03 It may need a FreeBSD config 2011-05-03T23:38:06 error: no hosts defaults found; please add 2011-05-03T23:39:03 Look for darwin.py or linux.py and copy to freebsd.py 2011-05-03T23:39:09 way ahead of you 2011-05-03T23:39:16 Then in defaults.py near the end 2011-05-03T23:40:10 i see, so they're overrides 2011-05-03T23:40:15 Yeap. 2011-05-03T23:40:43 It sort of follows the way rpmbuild works. 2011-05-03T23:40:53 But does not create a rpm. 2011-05-03T23:40:57 i wonder if i have an rtems.org email 2011-05-03T23:41:04 I do. 2011-05-03T23:41:05 i don't think we ever set that up, hmm 2011-05-03T23:41:11 Oh I mean address. 2011-05-03T23:41:24 oh, yeah i know, i was thinking about the license header 2011-05-03T23:41:34 Ah yes. 2011-05-03T23:41:38 Let me look 2011-05-03T23:42:21 not in aliases 2011-05-03T23:42:59 Can you send me the aliases file you would like 2011-05-03T23:43:04 sorry line 2011-05-03T23:44:08 i'll fix this up for freebsd 2011-05-03T23:44:11 thank you for using python :) 2011-05-03T23:44:45 *** monstr has joined #rtems 2011-05-03T23:47:08 Not a problem. It is a nice class that can parse spec files. 2011-05-03T23:58:35 kiwichris: the _host variable can be the output of ./config.guess right? 2011-05-04T00:06:49 kiwichris: so the purpose is to build everything and get tar files, then extract those to a prefix? 2011-05-04T00:06:56 Let me look 2011-05-04T00:07:02 Yes 2011-05-04T00:07:25 It does the build exactly the same way rpmbuild does on Linux for the yum tools 2011-05-04T00:07:36 Once set up you can stay in sync with head 2011-05-04T00:08:09 good god this is a lot of tools 2011-05-04T00:08:15 Yes it is the output of config.guess. 2011-05-04T00:08:23 my poor ssd will be filled up in a matter of hours 2011-05-04T00:09:09 Hmm config.guess could be added and run. I run config.sub and catch the output 2011-05-04T00:09:19 actually, hold on 2011-05-04T00:09:23 there are ports for all of these on freebsd 2011-05-04T00:09:33 Port for what ? 2011-05-04T00:09:45 /usr/ports/devel/i386-rtems-binutils /usr/ports/devel/i386-rtems-gcc /usr/ports/devel/i386-rtems-gdb 2011-05-04T00:09:53 for arm,i386,m68k,mips,powerpc,sh,sparc 2011-05-04T00:10:02 Are they current ? 2011-05-04T00:10:06 freebsd ports 2011-05-04T00:10:06 hem 2011-05-04T00:10:09 er, hrm 2011-05-04T00:10:26 gcc 4.5.2... 2011-05-04T00:10:36 rtems 4.11 2011-05-04T00:10:49 NIce. 2011-05-04T00:10:59 I think Ralf may build freebsd 2011-05-04T00:11:16 does ralf irc? i forget 2011-05-04T00:11:25 no 2011-05-04T00:12:10 so, lets say i use the ports versions for keeping this up to date 2011-05-04T00:12:18 that should be all i need to build the rtems package right? 2011-05-04T00:12:21 Sure. 2011-05-04T00:12:32 what do i use for a simulator? virtualbox ok? 2011-05-04T00:13:14 haha: ===> i386-rtems-gcc-4.5.2 is marked as broken: Does not link. 2011-05-04T00:13:15 sigh 2011-05-04T00:13:57 What version of FreeBSD are you on ? 2011-05-04T00:14:21 ports has no version but i'm on 9.0-CURRENT 2011-05-04T00:14:35 aparrantly the build is broken only when compiling i386 from amd64 2011-05-04T00:15:17 What about ub/rtems/linux/infrastructure/fedora/15/i386/i586-pc-freebsd8.2-* 2011-05-04T00:15:23 What is this ? 2011-05-04T00:15:43 i don't know what it is, what are you looking for? 2011-05-04T00:15:54 These look like packages so Ralf can build for FreeBSD on Fedroa. 2011-05-04T00:16:10 Under http::/www.rtems.org/ftp/.... 2011-05-04T00:16:50 ah 2011-05-04T00:17:34 and http://www.rtems.org/viewvc/rtems/contrib/crossrpms/freebsd8.2/ 2011-05-04T00:17:41 so I wonder where the output is. 2011-05-04T00:17:47 hmm 2011-05-04T00:18:57 i'll see if i can fix the port 2011-05-04T00:19:19 You will need SpecBuilder to get the magic 100% as RTEMS needs :) 2011-05-04T00:19:33 I would hack SpecBuilder to create the port. 2011-05-04T00:19:50 that won't work, i'll try using specbuilder first then 2011-05-04T00:20:01 Why not ? 2011-05-04T00:21:24 All SpecBuilder does is create a doit.sh shell script. 2011-05-04T00:21:34 This is all rpmbuilder does. 2011-05-04T00:22:08 *** projectgus has left #rtems 2011-05-04T00:22:56 beacuse port files are complicated 2011-05-04T00:23:05 it uses an entire framework it would not be possible to generate them 2011-05-04T00:23:07 Ever looked at spec files ? 2011-05-04T00:23:16 ever looked at a port? :) 2011-05-04T00:23:28 Yeap. 2011-05-04T00:23:46 http://www.freebsd.org/cgi/cvsweb.cgi/ports/devel/cross-gcc/Makefile?rev=1.18;content-type=text%2Fplain this is the one for the gcc cross building 2011-05-04T00:24:23 Yeap not that different. 2011-05-04T00:24:41 really? ports uses pmake 2011-05-04T00:25:05 Take a close look at spec.py. It contains many work arounds where the file has various hacks 2011-05-04T00:25:37 All I am saying is spec.py will parse a spec file. With that data you could write a port file. 2011-05-04T00:25:54 The complexity is in the contents of the spec file not the format. 2011-05-04T00:26:53 i'm trying spec-build now 2011-05-04T00:27:00 This complexity is what crossgcc needs and what Ralf does a fantastic job of maintaining. 2011-05-04T00:27:36 You can run each part stand alone. For example spec.py can run and dump a spec file. 2011-05-04T00:27:47 i'll try sometime maybe 2011-05-04T00:27:55 :) 2011-05-04T00:27:57 how will i run these tests once they're built? 2011-05-04T00:28:09 The RTEMS tests ? 2011-05-04T00:28:15 yeah 2011-05-04T00:28:30 With rtems-testing (which does not have my patch). 2011-05-04T00:28:35 I can send it. 2011-05-04T00:29:03 hmm yeah please do 2011-05-04T00:29:10 All you need is the sparc and i386 targets 2011-05-04T00:29:18 No need to build all of them. 2011-05-04T00:34:27 yeah i'm doing i386 for now 2011-05-04T00:34:49 Ok. 2011-05-04T00:35:16 Just sent my rtems-testing patch. It has been a while since I looked at it. 2011-05-04T00:37:01 nice thanks 2011-05-04T00:59:36 kiwichris: the wrong make is run in this, it uses 'make' 2011-05-04T00:59:42 on freebsd you need 'gmake' to get gnu make 2011-05-04T00:59:53 which isn't part of the base system you have to install it to get it 2011-05-04T01:00:00 In the platform file set gmake. 2011-05-04T01:00:10 i did it's ignored 2011-05-04T01:00:19 '__make': '/usr/local/bin/gmake', 2011-05-04T01:00:20 That is sad. 2011-05-04T01:00:28 it uses 'make' 2011-05-04T01:00:30 in the doit file 2011-05-04T01:01:04 Just looking now. 2011-05-04T01:02:02 Is it referencing /usr/bin/make ? 2011-05-04T01:02:05 download: http://www.rtems.org/ftp/pub/rtems/SOURCES/4.11/rtems-4.9.99.0-20090206-2.tar.bz2: failed: HTTP Error 404: Not Found 2011-05-04T01:02:08 no just 'make' 2011-05-04T01:02:20 Ok. 2011-05-04T01:02:47 Do not use the RTEMS spec file. 2011-05-04T01:02:47 that version seems rather old, why is it trying that? 2011-05-04T01:02:50 oh 2011-05-04T01:03:01 It is broken and should be removed or disabled. 2011-05-04T01:03:05 ah, i just saw 'cpukit' i didn't notice the rtems 2011-05-04T01:03:16 It allows a binary rtems release. 2011-05-04T01:03:26 gcc built now for gdb 2011-05-04T01:03:34 It is nice but messes with the development installs. 2011-05-04T01:04:15 Was the make problem just the cpukit one ? 2011-05-04T01:04:35 Nice. 2011-05-04T01:04:49 no 2011-05-04T01:04:52 all of them 2011-05-04T01:05:13 autoconf,automake,binutils happen to work using pmake (freebsd default 'make'), however gcc and gdb do not 2011-05-04T01:05:41 so SPECS/rtems-4.11-i386-rtems4.11-gcc.spec 2011-05-04T01:05:58 and gdb 2011-05-04T01:06:13 oh i see 2011-05-04T01:06:18 make is hardcoded in the spec files! 2011-05-04T01:06:22 Yeap. 2011-05-04T01:06:24 no wonder i didn't see it 2011-05-04T01:06:30 that will never work on freebsd :) 2011-05-04T01:06:54 i didn't think to look there 2011-05-04T01:08:46 Just looking into it. 2011-05-04T01:09:22 Been in the files for ever. 2011-05-04T01:09:28 I will ask Ralf. 2011-05-04T01:10:14 thanks 2011-05-04T01:10:18 i was going to open a ticket... 2011-05-04T01:12:40 Let me ask first. 2011-05-04T01:13:36 I suggest you hack the spec files and change make to %{__make} 2011-05-04T01:14:24 i just edited the doit file 2011-05-04T01:14:39 Sure. 2011-05-04T02:00:10 *** arvind_khadri has joined #rtems 2011-05-04T02:02:22 *** arvind_khadri has quit IRC 2011-05-04T02:02:43 *** arvind_khadri has joined #rtems 2011-05-04T02:49:01 *** qmor has joined #rtems 2011-05-04T02:52:25 hello to all 2011-05-04T02:53:22 i return with my problem with nvdisk again 2011-05-04T02:53:46 the mail list is rejecting my messages 2011-05-04T02:54:26 and i need example of using the nvdisk 2011-05-04T03:05:58 qmor, try http://www.rtems.org/ftp/pub/rtems/people/chrisj/file-system/rfs-20110315.tar.bz2 2011-05-04T03:14:33 thanks, but it not that i need 2011-05-04T03:14:52 i want to create filesystem in higher addresses of eeprom 2011-05-04T03:15:10 i have the functions to read, and to write the eeprom 2011-05-04T03:15:30 but in your example nvdisk creating in the heap 2011-05-04T03:15:46 and using the sram funct templates 2011-05-04T03:15:48 Ah ok. Hang on. 2011-05-04T03:16:09 That is it. You need to create a set of functions for your device. 2011-05-04T03:16:24 i have this func 2011-05-04T03:16:34 but i don't know how to use its 2011-05-04T03:17:00 const rtems_nvdisk_driver_handlers nv_io_ops = { 2011-05-04T03:17:00 eeprom_read, 2011-05-04T03:17:00 eeprom_write, 2011-05-04T03:17:00 eeprom_verify 2011-05-04T03:17:00 }; 2011-05-04T03:17:09 i saying about this 3 func 2011-05-04T03:17:33 Ok 2011-05-04T03:17:38 I will just have a look 2011-05-04T03:18:16 i'm trying to use its 2011-05-04T03:18:41 but when i'm debugging my app the functions not using 2011-05-04T03:18:48 i don't know why 2011-05-04T03:19:15 i could take you all of code via pasebin 2011-05-04T03:19:37 qmor, I am looking and doing a couple of other things at once :) 2011-05-04T03:20:21 qmor, what does not device descriptor look like ? 2011-05-04T03:20:28 qmor, what does your device descriptor look like ? 2011-05-04T03:20:46 rtems_nvdisk_device_desc rtems_nv_heap_device_descriptor[] = 2011-05-04T03:20:47 { 2011-05-04T03:20:47 { 2011-05-04T03:20:47 flags: 0, 2011-05-04T03:20:47 base: 0x7d000, 2011-05-04T03:20:47 size: 0x7d000, 2011-05-04T03:20:47 nv_ops: &nv_io_ops 2011-05-04T03:20:48 } 2011-05-04T03:20:48 }; 2011-05-04T03:21:06 sorry for pasting to channel directly 2011-05-04T03:21:21 And ns_io_ops is your functions table ? 2011-05-04T03:21:28 yes 2011-05-04T03:21:52 Then they should be called if you created the device correctly. 2011-05-04T03:22:12 yes, i think so 2011-05-04T03:22:28 I would turn the trace on and have a look. 2011-05-04T03:22:33 but it doesn't happened 2011-05-04T03:22:42 What does not happen ? 2011-05-04T03:22:47 The trace or the calls ? 2011-05-04T03:22:59 calls of my functions 2011-05-04T03:23:22 http://pastebin.com/cVuikdt2 2011-05-04T03:23:27 Does rtems_nvdisk_ioctl get called ? 2011-05-04T03:23:29 it's my app 2011-05-04T03:23:39 don't know 2011-05-04T03:23:49 testing just now 2011-05-04T03:26:02 yes, it is calling 2011-05-04T03:26:32 Great. I wonder why it is failing. I cannot see anything wrong with what you have. 2011-05-04T03:27:12 I have to head off for a while. I suggest you take a look at the ioctl function and see if you can see anything. 2011-05-04T03:27:22 It maybe confdefs.h related. 2011-05-04T03:27:35 eg not enough resources that maybe needed. 2011-05-04T03:27:39 cya 2011-05-04T03:34:58 *** xiangfu has quit IRC 2011-05-04T04:44:24 thanks 2011-05-04T04:44:43 i don't know why my functions doesn't calling 2011-05-04T04:45:15 i've write two pseudo func to rtems_driver_address_table for read and write 2011-05-04T04:45:49 and when i opening file and writing data via fwrite() nothing happens 2011-05-04T04:52:42 *** lj8175 has joined #rtems 2011-05-04T05:59:22 *** lekernel has quit IRC 2011-05-04T06:06:21 *** peerst has joined #rtems 2011-05-04T06:12:28 *** lekernel has joined #rtems 2011-05-04T07:01:17 *** arvind_khadri has quit IRC 2011-05-04T07:14:11 *** arvind_khadri has joined #rtems 2011-05-04T07:29:25 *** monstr has quit IRC 2011-05-04T07:34:54 *** dr__house has joined #rtems 2011-05-04T07:34:54 *** dr__house has joined #rtems 2011-05-04T08:00:33 *** zwj has joined #rtems 2011-05-04T08:04:44 *** qmor has quit IRC 2011-05-04T08:07:07 *** arvind_khadri has quit IRC 2011-05-04T08:11:10 *** sqm has joined #rtems 2011-05-04T09:08:21 *** gedare has joined #rtems 2011-05-04T09:47:11 *** gedare has quit IRC 2011-05-04T10:06:39 *** lj8175 has quit IRC 2011-05-04T10:07:14 *** sqm has quit IRC 2011-05-04T10:24:32 *** DrJoel has joined #rtems 2011-05-04T10:24:32 *** ChanServ sets mode: +o DrJoel 2011-05-04T10:25:30 *** DrJoel changes topic to "RTEMS.org and .com are back online. Still doing clean up and recovery at OAR so just let us know if you spot anything odd." 2011-05-04T10:27:14 *** dr__house has quit IRC 2011-05-04T10:27:20 *** dr__house has joined #rtems 2011-05-04T10:27:20 *** dr__house has joined #rtems 2011-05-04T10:29:13 *** dr__house has quit IRC 2011-05-04T10:29:19 *** dr__house has joined #rtems 2011-05-04T10:59:56 *** DrJoel has quit IRC 2011-05-04T12:31:00 *** dr__house has quit IRC 2011-05-04T15:05:28 hmm 2011-05-04T17:10:53 *** projectgus has joined #rtems 2011-05-04T18:06:37 *** projectgus has quit IRC 2011-05-04T18:10:01 *** projectgus has joined #rtems 2011-05-04T18:49:01 *** zwj has quit IRC 2011-05-04T19:11:27 kiwichris: around? 2011-05-04T19:11:32 yeap 2011-05-04T19:11:43 nice, i setup buildbot on rtbf32a 2011-05-04T19:11:51 it's running, along with a slave 2011-05-04T19:12:06 Neat and well done. 2011-05-04T19:12:09 i still need to setup slaves on our other build machines 2011-05-04T19:12:14 Cool. 2011-05-04T19:12:22 thanks but that was the easy part :) 2011-05-04T19:12:40 we need to go through how we want this to work, ie. section these into steps 2011-05-04T19:12:40 The hard is ? 2011-05-04T19:12:49 ok 2011-05-04T19:13:08 the toolchain is already on this machine, so our first step is to checkout rtems then run autogen and configure it 2011-05-04T19:13:12 after...? 2011-05-04T19:13:19 Can this take over from build bot ? 2011-05-04T19:13:36 do we need to build for each platform? do we use the same tree? 2011-05-04T19:13:55 Same tree and different output directories. 2011-05-04T19:14:10 ok, that's slightly tricky but not bad 2011-05-04T19:14:20 can we build for each platform+bsp combo? 2011-05-04T19:15:14 On rtbf32a look in /home2/work/chris/pq 2011-05-04T19:15:44 hmm no automounter is active :( 2011-05-04T19:15:52 oh sorry 2011-05-04T19:16:00 i typed /home2/chris/work hehe 2011-05-04T19:16:57 in pq/bin/pq-crontab is the build start 2011-05-04T19:18:53 hmm 2011-05-04T19:19:48 where is the cvs repo? 2011-05-04T19:20:50 on the phone 2011-05-04T19:20:52 ok 2011-05-04T19:21:52 nevermind found it 2011-05-04T19:29:04 kiwichris: feel free to answer when you're off the phone 2011-05-04T19:29:11 do we want to build for rtems 4.8,9,10,11? 2011-05-04T19:29:28 just 11 and may be 10 2011-05-04T19:30:51 ok 2011-05-04T19:31:28 PATH="/opt/rtems-4.11/bin/:$PATH" ./bootstrap 2011-05-04T19:31:31 should get me going right? 2011-05-04T19:31:45 yeap 2011-05-04T19:31:56 cool 2011-05-04T19:32:30 wow and i thought this was slow on my machine i guess an ssd makes a big difference 2011-05-04T19:33:51 kiwichris: are there any specific permutations of configure options we want to test? or is building vanilla platform+bsp combos ok? 2011-05-04T19:44:01 kiwichris: wow, re-running bootstrap per-platform/bsp is definatly out of the question 2011-05-04T19:44:44 It is not needed. 2011-05-04T19:44:56 Did my script pq-builder help ? 2011-05-04T19:45:12 i'm going through this by hand first to i can get a better understanding of what is required 2011-05-04T19:45:23 waiting for bootstrap to finish 2011-05-04T19:45:26 The configure flags is a big issue and if this could be solved it would be a huge win for everyone. 2011-05-04T19:45:38 Good idea. 2011-05-04T19:45:45 what is the issue? 2011-05-04T19:45:52 i can run as many permutations as required give me a list 2011-05-04T19:46:13 Developers and users only use some configure options and they are not tested in combinations. 2011-05-04T19:46:46 right, that is something every project has to deal with, for instance ntp uses 6 different configure lines per platform, which equates to 6 builds per platform 2011-05-04T19:46:50 Having a way to test this and get the results would make me more comfortable. I tend to discourage there use for this reason. 2011-05-04T19:46:53 i can do the same for rtems 2011-05-04T19:47:04 Fantastic. 2011-05-04T19:47:13 get me a list of what combinations you want 2011-05-04T19:47:17 that's trivial to do 2011-05-04T19:47:45 even if you get me a temp list we can modify it later, it will be a python dictionary 2011-05-04T19:47:54 I think --enable/disable-networking, c++ debug 2011-05-04T19:48:03 all we'd have to do is modify the dict and rehash the master and it'll work 2011-05-04T19:48:14 c++ debug? 2011-05-04T19:48:25 so --enable-networking, --disable-networking.. and? 2011-05-04T19:48:30 yeap 2011-05-04T19:48:46 can you give me the exact options? 2011-05-04T19:49:26 Just doing a configure --help now. 2011-05-04T19:49:30 ok 2011-05-04T19:50:15 --enable-multiprocessing 2011-05-04T19:50:16 enable multiprocessing interface 2011-05-04T19:50:16 --enable-posix enable posix interface 2011-05-04T19:50:16 --enable-networking enable TCP/IP stack 2011-05-04T19:50:16 --enable-cxx enable C++ support and build the rtems++ library 2011-05-04T19:50:16 --enable-tests enable tests (default:samples) 2011-05-04T19:50:18 --enable-rtems-debug enable RTEMS_DEBUG 2011-05-04T19:50:20 --enable-rtemsbsp="bsp1 bsp2 .." 2011-05-04T19:50:22 BSPs to include in build 2011-05-04T19:50:24 --enable-multilib build many library versions (default=no) 2011-05-04T19:50:26 --enable-docs enable building documentation (default:disabled) 2011-05-04T19:50:35 all those? 2011-05-04T19:51:07 so, for each of these options we want to build each platform+bsp combo? 2011-05-04T19:51:09 I suppose so. I am not sure about --enable-multiprocessing but the others yes. 2011-05-04T19:51:16 Yeap. 2011-05-04T19:51:27 That is max. 2011-05-04T19:51:43 Can we stage different runs ? 2011-05-04T19:51:49 well we don't have to build them all the time if we don't want to, we can build some after every commit 2011-05-04T19:51:53 For example everything once a week. 2011-05-04T19:51:57 and some after nth commit, or schedual it to run on certain days 2011-05-04T19:52:09 schedule too 2011-05-04T19:52:19 I would say a major build and package of the source as a snapshot once a week. 2011-05-04T19:52:27 With a report to a mailing list 2011-05-04T19:53:16 My build bot provides the source for each build each day. 2011-05-04T19:54:11 i'll most likely stagger them 2011-05-04T19:54:31 doing them all at once makes sense, there are certainly periods where developers aren't developing or pushing things to the repo 2011-05-04T19:54:41 i'll use those spaces to fire off a build etc 2011-05-04T19:54:50 Ok. 2011-05-04T19:54:54 the best solution is to get a new machine that continously builds them 2011-05-04T19:55:01 the sooner you know something has broken the better 2011-05-04T19:55:27 This machine can do this. It basically only runs my build bot and the CVS email. 2011-05-04T19:55:45 well probably what i'll endup doing is using ccache 2011-05-04T19:55:48 to cache the objects 2011-05-04T19:56:03 and running all these permutations immediatly, ccache can take the build down from hours to a few mins 2011-05-04T19:56:07 then once a week do a "from scratch" build 2011-05-04T19:56:29 Nice. I have used distcc. 2011-05-04T19:56:39 ccache caches object files 2011-05-04T19:57:00 i use it for all my development it's handy 2011-05-04T19:57:08 Given all the machines are kept up to date this should be fine. 2011-05-04T19:57:12 you can take full rebuilds from 10 mins down to about 4 seconds :) 2011-05-04T19:57:35 (depends on how much linking you do since that is not cached) 2011-05-04T19:57:57 So it is just for a single node ? 2011-05-04T19:58:11 it's an object cache 2011-05-04T19:58:13 it caches the .o 2011-05-04T19:58:30 I had heard of it but never looked into it. 2011-05-04T19:58:33 you run 'ccache cc ' and it caches the object ased on compiler flags and the timestamps/hash of includes and c files 2011-05-04T19:58:49 if it matches the next time you run it, it copies the .o out of the cache and doesn't compile it 2011-05-04T19:58:59 Neat. 2011-05-04T20:00:10 try it, go install ccache 2011-05-04T20:00:15 CC="ccache cc" 2011-05-04T20:00:24 I will. 2011-05-04T20:00:25 do a build to load your cache 2011-05-04T20:00:33 make distclean to nuke everything, reconfig and type make 2011-05-04T20:00:37 you'll be surprised :) 2011-05-04T20:03:38 *** xiangfu has joined #rtems 2011-05-04T20:13:32 kiwichris: do you have a list of platform+bsp in chrisj/pq? 2011-05-04T20:13:58 I will look. 2011-05-04T20:14:32 and i can't see how i'm supposed to build for a specific platform and bsp 2011-05-04T20:14:42 Line 106 in bin/pq-builder 2011-05-04T20:14:52 Gets the list of targets from the tools installed. 2011-05-04T20:15:13 that's an interesting way to do it.. 2011-05-04T20:15:14 Means we adjust to the users set of tools. 2011-05-04T20:15:47 i see RTEMS_BSP=x as a make option 2011-05-04T20:15:52 this will build that specific bsp? 2011-05-04T20:15:59 Yes. 2011-05-04T20:16:15 how do i target the platform? or is each BSP unique so that is not required? 2011-05-04T20:16:15 I think I use --enable-rtems-bsp=.... 2011-05-04T20:16:44 We use target+bsp rather than platform+target 2011-05-04T20:16:51 Just looking. 2011-05-04T20:17:20 oh, so i have to re-configure to target a specific bsp? 2011-05-04T20:17:33 so --target=target --enable-rtems-bsp=bsp 2011-05-04T20:17:38 Not sure. I am just looking now. 2011-05-04T20:17:39 then make; make install will get me what i need? 2011-05-04T20:17:41 ok thanks 2011-05-04T20:17:55 At line 260 I think I detect the targets from the dirs created. 2011-05-04T20:18:25 I do a build per BSP in a single pass and get the results. 2011-05-04T20:18:59 Via make. 2011-05-04T20:19:25 I suspect using configure or make performance will not make a great deal of difference. The major part of configure runs under make. 2011-05-04T20:19:27 yeah doing that with buildbot is pointless, i'll create one builder per target+bsp 2011-05-04T20:19:36 then clone those for each configure permutation 2011-05-04T20:19:45 Great. 2011-05-04T20:20:10 I am not sure what is the best solution for getting the list of BSPs. 2011-05-04T20:20:36 I will ask Ralf. 2011-05-04T20:20:47 http://www.rtems.org/ftp/pub/rtems/people/chrisj/pq/html/ 2011-05-04T20:20:50 they're all on here right? 2011-05-04T20:21:18 Yes. I do a single configure for a target and then check the directory for valid Makefiles. 2011-05-04T20:21:40 ohh i see 2011-05-04T20:21:42 This works because I looked and hacked a solution. Ralf could change the layout and I would break. 2011-05-04T20:21:50 well i need a static list so i'll just scrape it from the website 2011-05-04T20:21:54 We should really ask configure for the list. 2011-05-04T20:22:05 buildbot will break too if he does that 2011-05-04T20:22:38 I would prefer an automatic way so adding a new BSP does not break the bot or get missde. 2011-05-04T20:22:46 it won't break the bot if we add one 2011-05-04T20:22:51 it won't get built until we add it 2011-05-04T20:23:12 I will ask and get back to you on this one. 2011-05-04T20:23:18 you should not dynamically add/remove builders based on the current source coming in, too much can go wrong 2011-05-04T20:23:43 Ok so the list is static. No problem. 2011-05-04T20:23:51 you need to test based on static configurations because you're trying to ensure *that* specific config works 2011-05-04T20:23:56 we can make it easier of course 2011-05-04T20:24:06 i can make a web app or something and we can dynamically load the list from the database 2011-05-04T20:24:15 The configure repors the list of BSPs. 2011-05-04T20:24:28 Ok. 2011-05-04T20:24:33 we'll figure out a solution that works out for everyone i'm not worried about it 2011-05-04T20:24:40 for now i'll get it working static and we'll sort out what we need after 2011-05-04T20:24:45 Then neither am I :) 2011-05-04T20:24:55 i'm going to assume that ralph doesn't change this daily? :) 2011-05-04T20:25:09 It is a major change. 2011-05-04T20:25:26 If the list is static all this does not matter. 2011-05-04T20:25:51 yeah for now it will be a static list in the config file for the master 2011-05-04T20:44:25 1,098 builders, not bad 2011-05-04T20:44:37 we're going to need more hardware :) 2011-05-04T20:44:41 Oooo nice 2011-05-04T20:44:54 I will Joel know. 2011-05-04T20:44:58 i'll have something pretty to show you in a moment 2011-05-04T20:55:31 oh no wonder this is so slow builds are running 2011-05-04T21:15:52 *** xiangfu has quit IRC 2011-05-04T21:16:18 *** xiangfu has joined #rtems 2011-05-04T21:57:59 verm__, back 2011-05-04T22:15:24 same just sat down 2011-05-04T22:35:02 kiwichris: wow this machine sure slows down a lot when builds are going on 2011-05-04T22:35:18 Memory an issue ? 2011-05-04T22:35:31 It is only running 32bit 2011-05-04T22:36:02 yeah memory looks like it's an issue but also disk contention 2011-05-04T22:36:30 kiwichris: why 32bit anyway? surely redhat supports using 32bit binaries 2011-05-04T22:36:35 was it nessicary to make the entire os 32bit? 2011-05-04T22:36:53 It was an old machine before that died so it was moved. 2011-05-04T22:37:20 The 32bit was for testing 32bit builds of the tools if we had a problem. 2011-05-04T22:39:14 what i mean is, you can install 32bit tools on a 64bit platform 2011-05-04T22:39:34 you could have chrooted it, several options were available 2011-05-04T22:39:55 Yeah but it was an old box. It is now not as old. 2011-05-04T22:40:18 oh, you just moved the drive? 2011-05-04T22:40:28 I think so. 2011-05-04T22:40:35 ah, bummer 2011-05-04T22:40:42 Not sure. Joel and Jeff did it. 2011-05-04T22:43:55 how many other build machines do we have other than rtbf32a? 2011-05-04T22:45:46 Just looking 2011-05-04T22:46:49 cool can you tell me the arch,cpu,memory i don't have access to them all.. 2011-05-04T22:48:31 Cannot find the page 2011-05-04T22:51:16 ok no worries 2011-05-04T22:52:30 PATH="/opt/rtems-4.11/bin:$PATH" /home/amar/src/rtems/configure --target="i386-rtems4.11" 2011-05-04T22:52:34 there is nothing wrong with this configure line right? 2011-05-04T22:52:46 gmake RTEMS_BSP="pc686" 2011-05-04T22:52:48 and that line should work 2011-05-04T22:52:49 http://www.rtems.org/munin/ 2011-05-04T22:52:51 er 'make' that is 2011-05-04T22:53:23 Apart from building over NFS via the automounted home. 2011-05-04T22:53:28 try /home2 2011-05-04T22:53:51 No need for gmake on Linux :) 2011-05-04T22:54:06 hmm it's on nfs? i asked jeff to make sure it was not on nfs 2011-05-04T22:54:20 Your home ? 2011-05-04T22:54:26 yeah 2011-05-04T22:54:42 Ah ok. 2011-05-04T22:54:44 yeah it's not 2011-05-04T22:55:17 You are correct. You have a special entry in /etc/passwd and not from NIS 2011-05-04T22:57:03 in any case, that configure line does not work, when i try to build i get errors 2011-05-04T22:57:07 gmake[2]: Entering directory `/home/amar/src/config/i386-rtems4.11/c/pc686' 2011-05-04T22:57:08 gmake[2]: *** No rule to make target `all'. Stop. 2011-05-04T22:57:55 Did you bootstrap with the path set ? 2011-05-04T22:58:04 i did 2011-05-04T22:58:44 Are you building in the source tree ? 2011-05-04T22:58:52 no, it won't let you do that 2011-05-04T22:58:58 Hmmm 2011-05-04T22:59:06 well it doesn't let you configure there i mean 2011-05-04T22:59:13 Good. 2011-05-04T22:59:15 but i bootstrapped my source with the path and used the configure line above 2011-05-04T22:59:21 Yes this is correct. 2011-05-04T22:59:59 hmm if i type 'make all' it does something 2011-05-04T23:00:09 it starts configuring 2011-05-04T23:00:21 but if i do make RTEMS_BSP=xxx it just tries building and fails 2011-05-04T23:01:38 Something is broken in the configure boot strap. 2011-05-04T23:02:01 head c/config.log you have --target=i686-pc-linux-gnu 2011-05-04T23:02:13 i didn't set that 2011-05-04T23:02:18 I think it should RTEMS. 2011-05-04T23:02:20 I know 2011-05-04T23:02:26 i just ran PATH="/opt/rtems-4.11/bin:$PATH" ./bootstrap 2011-05-04T23:02:28 ah 2011-05-04T23:03:00 if you can tell me what date is safe to checkout i'll rewind so i can get this working :) 2011-05-04T23:03:15 No idea. 2011-05-04T23:03:24 This normal is never broken 2011-05-04T23:04:00 configure:4596: running /bin/sh '/home/amar/src/rtems/c/configure' '--prefix=/opt/rtems-4.11' '--host=i386-rtems4.11' '--build=i686-pc-linux-gnu' '--target=i386-rtems4.11' '--with-target-subdir=i386-rtems4.11' '--exec-prefix=/opt/rtems-4.11/i386-rtems4.11' '--includedir=/opt/rtems-4.11/i386-rtems4.11/include' '--cache-file=/dev/null' '--srcdir=/home/amar/src/rtems/c' 2011-05-04T23:04:16 Notice the target is pass correctly 2011-05-04T23:04:22 right 2011-05-04T23:04:33 i wonder if this is why it was broken for me at home too 2011-05-04T23:04:57 Could be 2011-05-04T23:05:05 I am working with my builder. 2011-05-04T23:05:17 It would not have got as far as it is. 2011-05-04T23:05:34 hmm 2011-05-04T23:05:42 Can I su to you and play ? 2011-05-04T23:05:52 absolutely, please do 2011-05-04T23:05:58 This is all cli driven ? 2011-05-04T23:06:10 what is? 2011-05-04T23:06:18 oh, i've been doing this all by hand yes 2011-05-04T23:06:19 This build ? 2011-05-04T23:06:32 it's always better to figure it out by hand so i can properly craft an automated process 2011-05-04T23:06:42 Great 2011-05-04T23:06:47 if you want to write a small sh script that's fine too 2011-05-04T23:08:33 Ok. Just running a bootstrap now. 2011-05-04T23:10:39 ok, what did you use? 2011-05-04T23:10:47 CL that is 2011-05-04T23:11:14 export PATH=/opt/rtems-4.11/bin:$PATH 2011-05-04T23:11:19 ./bootstrap -c 2011-05-04T23:11:22 ./bootstrap 2011-05-04T23:11:28 Separate lines 2011-05-04T23:15:52 hmm that would be very odd of exporting it made any difference 2011-05-04T23:18:28 Still bootstraping 2011-05-04T23:19:15 *** monstr has joined #rtems 2011-05-04T23:25:01 yeah i know it takes forever 2011-05-04T23:25:54 kiwichris: rtems is a good candidate to use waf 2011-05-04T23:25:59 it would make the build a lot more sane 2011-05-04T23:26:04 what is waf ? 2011-05-04T23:26:15 http://code.google.com/p/waf/ 2011-05-04T23:26:27 it's a build system written in python it's really good 2011-05-04T23:28:40 it would be a big task though 2011-05-04T23:29:14 it also supports being shipped as a single file, so we can put it into our distfiles all a user needs is python 2011-05-04T23:29:22 on the phone 2011-05-04T23:29:26 yep no worries 2011-05-05T00:45:28 *** xiangfu has quit IRC 2011-05-05T00:54:03 *** sqm has joined #rtems 2011-05-05T01:07:25 verm__, it is building. 2011-05-05T01:07:38 what the. 2011-05-05T01:07:49 so RTEMS_BSP works? 2011-05-05T01:08:16 I just cd to the config dir ran configure and make 2011-05-05T01:09:15 are you building right now? 2011-05-05T01:09:21 Yeap 2011-05-05T01:09:28 can you stop so i can try? 2011-05-05T01:09:39 Done 2011-05-05T01:09:53 yeah yours looks way different than mine 2011-05-05T01:10:17 I wonder if the PATH at the start does something 2011-05-05T01:10:25 so you ran configure --target=i386-rtems4.11 2011-05-05T01:10:33 yes 2011-05-05T01:10:36 having the path at the start sets it in your environment the same way export does 2011-05-05T01:10:49 I just coped and pasted yours from confi.log 2011-05-05T01:10:53 I just coped and pasted yours from config.log 2011-05-05T01:11:29 Yes I know 2011-05-05T01:12:07 hmm it is working 2011-05-05T01:12:12 that's messed up thank you for checking 2011-05-05T01:12:47 No problem. Maybe you had bootstrapped with the wrong path and did it again with the correct path without a bootstrap -c 2011-05-05T01:12:50 to clean it 2011-05-05T01:12:53 kiwichris: so, i'll want to configure for each target 2011-05-05T01:12:57 then do one build per BSP 2011-05-05T01:13:18 kiwichris: that's probably what i did 2011-05-05T01:13:24 i didn't know i needed a -c in between 2011-05-05T01:13:25 For each buildslave I assume it is a separate build ? 2011-05-05T01:13:44 Only when the versions get stuffed up of autoconf 2011-05-05T01:14:08 161 ./bootstrap 2011-05-05T01:14:08 169 PATH="/opt/rtems-4.11/bin/:$PATH" ./bootstrap 2011-05-05T01:14:10 exactly what i did 2011-05-05T01:14:25 i remember now because the first one failed due to an incorrect autoconf 2011-05-05T01:15:25 Nice. That solves that one. 2011-05-05T01:17:05 yeah, thanks.. now i need to figure out the best way to do this 2011-05-05T01:17:14 Ok 2011-05-05T01:17:28 running bootstrap more than once is not an option since it takes too long 2011-05-05T01:17:33 Is each build in a clean workspace ? 2011-05-05T01:17:45 so that means an initial step to create a bootstrapped snapshot 2011-05-05T01:17:52 Yes 2011-05-05T01:17:52 kiwichris: it doesn't have to be 2011-05-05T01:18:47 after the snapshot is made i'll configure for each target 2011-05-05T01:19:02 after a target is configured, that'll trigger the bsps to build in order 2011-05-05T01:19:04 for that target 2011-05-05T01:19:06 Yes. This is what I do. 2011-05-05T01:19:42 so snapshot -> [targets] -> [bsps] the entire chain needs to be locked too 2011-05-05T01:19:58 Yes. 2011-05-05T01:20:13 we can't start rebuilding a target until we know no more bsps will be built for that target 2011-05-05T01:20:14 hmmm 2011-05-05T01:20:41 there's also the issue of testing, does testing happen out-of-band? 2011-05-05T01:20:41 ? 2011-05-05T01:20:55 can i install the resulting test binaries somewhere? 2011-05-05T01:21:12 or do we want to run the tests right after the bsp has been built? (this is fine) 2011-05-05T01:21:20 Yes but why. Why not test in the build tree ? 2011-05-05T01:21:25 we can 2011-05-05T01:21:35 sometimes people want to split the building and testing i don't care which we do 2011-05-05T01:21:46 Yes. If you can get a list (ie current 'find . -name \*.exe') 2011-05-05T01:22:06 it'll be a static list i'll write a python script to execute them 2011-05-05T01:22:13 Ok. 2011-05-05T01:22:19 i want to track which work and which don't as well as the line-level pass/fail 2011-05-05T01:22:27 Ok 2011-05-05T01:22:30 kiwichris: how much code is common between bsps for a target? 2011-05-05T01:22:47 c/src/lib/libbsp/shared 2011-05-05T01:22:56 cpukit 2011-05-05T01:23:01 c/src/libcpu 2011-05-05T01:23:05 c/src/libchip 2011-05-05T01:23:14 Lots :) 2011-05-05T01:23:59 oh, hold on 2011-05-05T01:24:05 so it recompiles each of these per-bsp? 2011-05-05T01:24:11 yes 2011-05-05T01:24:19 Everything can be differrent. 2011-05-05T01:24:37 For example a BSP maybe a m68k but a cpu32 or a MCF5235 2011-05-05T01:24:43 *** sqm has quit IRC 2011-05-05T01:24:48 ahh ok that's good 2011-05-05T01:24:53 then i'll re-configure for each bsp 2011-05-05T01:24:58 that solves a major tricky spot 2011-05-05T01:25:03 And beyond the gcc -march -mcpu flags there is another level is control within RTEMS. 2011-05-05T01:25:06 rather than doing configure once-per-target 2011-05-05T01:25:39 running configure doesn't take very long, skipping it won't save us much time 2011-05-05T01:25:56 and this also tests to make sure that configure works when setting that specific bsp during configure-time 2011-05-05T01:25:58 The --enable-multilb is an option that causes the common code to be built to the gcc multilib configuration. This is a more common approach 2011-05-05T01:26:15 Correct. A lot of configure happens in the make phase anyway 2011-05-05T01:26:27 alright i'll do it this way 2011-05-05T01:26:33 Ok 2011-05-05T01:26:39 that'll remove the need for any locking 2011-05-05T01:27:02 so bsps can be built for any revision and we can split them across machines 2011-05-05T01:27:17 before we would have had to build all ARM bsps on the same machine since they'd be sharing the source directory 2011-05-05T01:27:22 if we re-run configure that won't matter 2011-05-05T01:27:31 Ok make sense. 2011-05-05T01:27:41 keeping builds compartmentalisedpast is the best method 2011-05-05T01:27:53 since all we have to do is "throw hardware" at the builds if we want them to be faster 2011-05-05T01:28:06 Sure 2011-05-05T01:28:45 if we have to start grouping builds on machines it can make things annoying since some targets will eat more cpu than others.. since there are a variable amount of bsps it would have been a lot more work to balance it 2011-05-05T01:28:46 nice, ok 2011-05-05T01:29:35 yeah that would have been fun to get correct. When Ralf spins tools he works all the hardware to its limits 2011-05-05T01:30:27 i should have this at least building for each bsp tonight (vanilla) 2011-05-05T01:30:34 Nice. 2011-05-05T01:30:40 i wish bootstrapping didn't take so long :( 2011-05-05T01:30:49 12 mins between each test is going to be painful 2011-05-05T01:31:02 In the pq is the Python code that parses the output to gather all the errors and warning. 2011-05-05T01:31:15 It that of any use ? 2011-05-05T01:31:18 buildbot can do that 2011-05-05T01:31:25 not automatically 2011-05-05T01:31:34 i can capture stdout and put it into a seperate log 2011-05-05T01:31:38 This way I do not need to build and we can see the warnings etc 2011-05-05T01:31:44 Ok 2011-05-05T01:31:53 oh i see how you've done it 2011-05-05T01:32:00 yeah i can re-use that script right inside buildbot 2011-05-05T01:32:04 where is it? 2011-05-05T01:32:05 Nice 2011-05-05T01:32:24 "/home2/work/chris/pq" 2011-05-05T01:32:35 then in bin. 2011-05-05T01:32:37 how many hours does a build of all bsps take on this machine? 2011-05-05T01:32:56 Its on the that pages I create 2011-05-05T01:32:59 hang on 2011-05-05T01:33:19 I start at 0:20 2011-05-05T01:34:02 4hrs for the normal build 2011-05-05T01:34:08 where is the script that creates: http://www.rtems.org/ftp/pub/rtems/people/chrisj/pq/html/h8300-h8sim-20110426-1.html 2011-05-05T01:34:16 'normal'? this is all 122 bsps? that's not bad 2011-05-05T01:34:25 thens about 8hr for the multilib option 2011-05-05T01:35:18 not too bad 2011-05-05T01:35:38 i'll have to ask joel what the other machines are doing and how much spare cpu we have 2011-05-05T01:35:45 we can do a variable amount of concurrent builds on each machine 2011-05-05T01:35:45 With an 8 core box it would drop quickly 2011-05-05T01:36:00 rather than doing make -j 8 2011-05-05T01:36:05 The load depends on Ralf. 2011-05-05T01:36:07 i'd do 6 or so concurrent builds, maybe 7 2011-05-05T01:36:18 Yes 2011-05-05T01:36:24 does ralf use all the machines? 2011-05-05T01:36:45 Not sure what he is using but the tools are in the order 24hr+ 2011-05-05T01:44:56 hmm 2011-05-05T01:45:04 well i know joel was talking about buying some new hardware the other week 2011-05-05T01:45:28 Is there a way for Ralf to take a machine out while he uses it ? 2011-05-05T01:46:36 stopping the slave is probably the best way 2011-05-05T01:46:54 the best solution is to dedicate machines to buildbot only and nothing else 2011-05-05T01:47:10 Ok. Joel will have to comment on this. 2011-05-05T01:47:26 since it's an automated process it's better nothing outside interfere with it, we can work out solutions as nessicary though it's not a problem 2011-05-05T01:48:06 i can self-manage load, ie pass builds to the next slave if the current one is at too high of a load 2011-05-05T01:48:33 Ok. So if Ralf starts a build the bot will see a load and move on ? 2011-05-05T01:48:36 it's a good way to make sure they're always in use, but not friendly to anyone else who wants to use the machine :) 2011-05-05T01:48:53 yeah, but if it's already building it will take a little time for it to notice and start skipping 2011-05-05T01:49:01 Sure. 2011-05-05T01:49:05 if buildbot is already building that is, he'll probably get annoyed by it 2011-05-05T01:49:10 i know i would 2011-05-05T01:49:24 This may just be how it is. We do not have endless resources. 2011-05-05T01:50:04 When does it look ? At the end of a BSP build ? 2011-05-05T01:51:18 well it'll only look when something tries to build there 2011-05-05T01:51:28 any builds currently running will have to finish 2011-05-05T01:51:46 And that unit is what, ie bsp ? 2011-05-05T01:52:00 yeah one build will be one bsp 2011-05-05T01:52:15 That is not much, ie 4hr/122 2011-05-05T01:52:45 for one box it's not 2011-05-05T01:53:11 depending on what our budget is for hardware we could take it down to less than 30 mins or so i think 2011-05-05T01:53:27 Nice. 2011-05-05T01:53:27 that's just for building how long do the tests usually take? 2011-05-05T01:53:44 No idea. They can take a while because some have to timeout. 2011-05-05T01:53:52 less than 30 mins for a nominal price that is, if we have a good budget we could do it in < 5 mins :) 2011-05-05T01:53:58 The simulators are not perfect 2011-05-05T01:54:12 timeout? what do you mean? 2011-05-05T01:54:28 The simulators do not exit cleanly 2011-05-05T01:54:46 that's brutal, why do we want to wait for it to time out? 2011-05-05T01:54:53 we can't convince it to exit earlier? 2011-05-05T01:55:02 You will have to ask Joel. This is his area. 2011-05-05T01:55:04 if we know the test has run to it's end 2011-05-05T01:55:05 ok 2011-05-05T01:55:12 and what rtems-testing contains. 2011-05-05T01:55:18 *** sqm has joined #rtems 2011-05-05T01:55:40 We do not know what currently because we capture the output and do not process it I think 2011-05-05T01:56:09 Just because a program exits the CPU does not stop and it is a simulator 2011-05-05T01:59:21 right, i understand that part, what i was saying is as soon as we get the output we need from our test we should be able to stop the simulator right after, rather than waiting for it 2011-05-05T01:59:42 there might be a reason, i dunno -- but waiting for timeouts would not be ideal. 2011-05-05T01:59:59 We would need to change the way rtems-testing works. It would be nice to to but it has not been a priority. 2011-05-05T02:00:26 I think there is some complexity. A python wrapper would help. Check out execute.py in the SpecBuilder. 2011-05-05T02:00:29 *** projectgus has quit IRC 2011-05-05T02:01:47 *** sqm has quit IRC 2011-05-05T02:03:45 yeah i can work something into my python script 2011-05-05T02:03:47 which reminds me 2011-05-05T02:03:55 kiwichris: how can i actaully run the tests? 2011-05-05T02:04:18 Did my patch apply to rtems-testing ? 2011-05-05T02:04:41 i havn't tried it yet, there is nothing setup on rtbf32a? 2011-05-05T02:04:49 No. 2011-05-05T02:05:00 Check out rtems-testing and apply my patch. 2011-05-05T02:05:18 Then somewhere do a configure, make, make install. 2011-05-05T02:05:48 For me and my testing I use a different prefix, eg ~/rtems/4.11-testing 2011-05-05T02:06:03 Then put this at the start of the path. 2011-05-05T02:06:10 You should then have rtems-testing. 2011-05-05T02:06:20 I mean rtems-test 2011-05-05T02:06:31 Then you can rtems-test sis executable 2011-05-05T02:06:42 where sis is the specific BSP. 2011-05-05T02:07:05 I think rtems-test as an option to list all the BSPs that have simulator support. 2011-05-05T02:07:22 The SIS is a nice one to play with because the simulator is built into the sparc gdb 2011-05-05T02:07:43 Other such as qemu need extra packages and support such as boot disk images 2011-05-05T02:08:19 *** sqm has joined #rtems 2011-05-05T02:08:21 ah 2011-05-05T02:08:22 Joel knows all about this. I know he has machine that do all the testing 2011-05-05T02:09:42 hmm 2011-05-05T02:10:43 My patch cleans up rtems-testing so the interface is a single command rather than a command per BSP/simualtor 2011-05-05T02:11:10 do i have to bootstrap before patching? 2011-05-05T02:11:55 No. This is very different. 2011-05-05T02:12:00 hmmm 2011-05-05T02:12:04 why didn't that patch apply cleanly 2011-05-05T02:12:19 rtems testing is a separate CVS module to rtems 2011-05-05T02:12:22 oh now it did 2011-05-05T02:12:26 yeah i know 2011-05-05T02:12:28 it failed the first time 2011-05-05T02:12:38 ok now it applied fine 2011-05-05T02:12:41 so i run bootstrap first? 2011-05-05T02:12:44 It may have changed and broken the patch. 2011-05-05T02:12:46 yes 2011-05-05T02:12:59 should i set the path to /opt/rtems4.11/bin? 2011-05-05T02:13:04 No 2011-05-05T02:13:07 It is all host code 2011-05-05T02:13:09 ok, running 2011-05-05T02:13:34 It build Joel's coverage analysis tool. That is a whole other world of testing :) 2011-05-05T02:15:31 yeah i had long discussions with joel about that at the mentors summit last year 2011-05-05T02:25:49 kiwichris: does this need to be installed anywhere? 2011-05-05T02:27:08 Yes. You need to set a prefix on configure. 2011-05-05T02:27:28 I tend to use something like ~/rtems/4.11-testing 2011-05-05T02:35:40 cool, well it's installed 2011-05-05T02:36:03 Do you have rtems-test ? 2011-05-05T02:37:45 the binary? yes 2011-05-05T02:37:57 shell script ? 2011-05-05T02:38:02 yeah it's a shell script 2011-05-05T02:38:35 Best source of doco is the script :( 2011-05-05T02:38:46 ah hehe ok 2011-05-05T02:39:13 rtems-test sis c/xxxx/samples/ticker/ticker.exe 2011-05-05T02:39:18 for a SIS BSP 2011-05-05T02:39:25 which is a sparc device 2011-05-05T02:40:20 If you look under $PREFIX/libexec/rtems/testing/sim-scripts you will find all the scripts 2011-05-05T02:40:45 These are all Joel's work and rather impressive. 2011-05-05T02:44:14 nice 2011-05-05T02:44:37 building sparc-sis now 2011-05-05T02:50:02 The rtems-test script is sort of Joel's things. I do not mind it changing lots and even being in Python. 2011-05-05T02:54:17 how do i build the testsuite for sis now? 2011-05-05T02:54:43 Oh yeah this one --enable-tests on the configure of RTEMS. 2011-05-05T02:54:52 You should have the samples to try the rtems-test 2011-05-05T02:55:03 yeah it's probably worth it revisit all of this and bring it under a common infastructure 2011-05-05T02:55:09 There are options to --enable-tests 2011-05-05T02:55:10 oh i have to enable them 2011-05-05T02:55:16 kiwichris: yep there are some 2011-05-05T02:55:35 All the tests really crank the build time up. 2011-05-05T02:56:06 I wonder if I have them enabled on the builder ... hmmm... I thought I did. 2011-05-05T02:57:53 You can --enable-tests[=yes/no] or --enable-tests=samples 2011-05-05T02:57:55 so just --enable-tests do i have to select a different target? 2011-05-05T02:58:05 No 2011-05-05T02:58:17 err make target that is not build target 2011-05-05T02:58:23 ok it's the =yes i needed 2011-05-05T02:58:32 It is a configure option 2011-05-05T02:58:54 hmm --enable-tests=yes or --enable-tests just gives me the samples 2011-05-05T02:59:29 PATH=/opt/rtems-4.11/bin:$PATH /home/amar/src/rtems/configure --target=sparc-rtems4.11 --prefix=/home/amar/prefix-rtems --enable-tests 2011-05-05T02:59:30 make RTEMS_BSP="sis" 2011-05-05T02:59:30 is the RTEMS source the macro is in aclocal/enable_tests.m4 2011-05-05T03:00:29 Hmm I do not understand then 2011-05-05T03:00:37 maybe reconfiguring doesn't work 2011-05-05T03:00:49 i wonder if i need to wipe then rebuild everything 2011-05-05T03:00:59 it might not be picking up that i enabled them 2011-05-05T03:01:01 i'll try that 2011-05-05T03:01:02 I do not know. You may. 2011-05-05T03:01:21 I will send you a script I use. 2011-05-05T03:04:09 yeah that was it, wiping and restarting did it 2011-05-05T03:04:14 now it's building all the tests 2011-05-05T03:04:37 thank you for the script 2011-05-05T03:05:04 It makes doing this simpler. It is also why I never remember the configure options :) 2011-05-05T03:06:52 heh 2011-05-05T03:07:20 Every time I get ask I need to head config.log 2011-05-05T03:08:00 i know how that can be.. 2011-05-05T03:08:22 Did you see we had ntpd issues on rtems.org on power up ? 2011-05-05T03:08:31 nope 2011-05-05T03:08:42 I just send the email about it to you. 2011-05-05T03:08:46 clock was way out of sync? 2011-05-05T03:08:53 I suspect so. 2011-05-05T03:09:25 well it's not an ntpd issue, the problem was that it was disabled 2011-05-05T03:09:36 Oh. 2011-05-05T03:09:36 i'm guessing the clock is broken and that was causing the problems, ntpd will take care of that 2011-05-05T03:09:53 Ok. 2011-05-05T03:09:59 I wonder why it was disabled. 2011-05-05T03:10:03 Must go 2011-05-05T03:10:06 someone probably forgot to enable it 2011-05-05T03:15:30 *** xiangfu has joined #rtems 2011-05-05T03:35:00 kiwichris: cvs is painful :( 2011-05-05T04:19:13 kiwichris: hmm there's a lot of linking during the build and ccache wasn't used for every compile (most but not all) the time before was 18r, 9u, 3s (mins) and after was 8r, 2u, 3s (real/user/sys) 2011-05-05T04:32:33 *** sqm has quit IRC 2011-05-05T04:37:28 *** sqm has joined #rtems 2011-05-05T04:49:22 hmmm 2011-05-05T04:49:30 i think the first task is going to be setting up automatic cvs -> git mirroring 2011-05-05T05:02:55 *** sqm has quit IRC 2011-05-05T05:03:12 *** sqm has joined #rtems 2011-05-05T05:09:51 *** sqm has quit IRC 2011-05-05T05:10:36 *** sqm has joined #rtems 2011-05-05T05:29:41 *** xiangfu has quit IRC 2011-05-05T05:29:46 *** sqm has quit IRC 2011-05-05T05:37:12 *** sqm has joined #rtems 2011-05-05T05:43:52 *** sqm has quit IRC 2011-05-05T05:44:50 *** sqm has joined #rtems 2011-05-05T06:59:19 *** lekernel has quit IRC 2011-05-05T07:12:17 *** lekernel has joined #rtems 2011-05-05T07:51:44 *** lj8175 has joined #rtems 2011-05-05T07:52:30 *** sqm has quit IRC 2011-05-05T07:52:38 kiwichris: around? 2011-05-05T07:58:01 *** sqm has joined #rtems 2011-05-05T08:06:54 *** xiangfu has joined #rtems 2011-05-05T08:12:35 *** xiangfu has quit IRC 2011-05-05T08:22:54 *** xiangfu has joined #rtems 2011-05-05T08:35:10 *** monstr has quit IRC 2011-05-05T08:37:38 *** zwj has joined #rtems 2011-05-05T09:05:14 *** DrJoel has joined #rtems 2011-05-05T09:05:14 *** ChanServ sets mode: +o DrJoel 2011-05-05T09:14:50 hi Joel, you are back 2011-05-05T09:17:03 first real day sitting in front of a computer since last Monday. 2011-05-05T09:17:20 If you saw my blog, I actually finished teaching the class outside without power. LOL 2011-05-05T09:18:55 lol, welcome back :) 2011-05-05T09:23:49 *** sqm has quit IRC 2011-05-05T09:30:03 *** sqm has joined #rtems 2011-05-05T09:37:03 I feel very behind on GSOC. We need to settle mentors and co-mentors and should have done it by now. 2011-05-05T09:37:23 Also should have had dedicated mailing list to make announcements ready .. soon.. after mentors are settled. 2011-05-05T09:49:33 lj8175: thanks for help but i meant changing ip address after bootup.. :) like switch_ip_address(new_ip); 2011-05-05T09:49:47 DrJoel: wb ! :) 2011-05-05T09:50:48 * DrJoel is happy to be here. it has been a LONG week. But others have it worse 2011-05-05T09:50:50 i read the blog.. you have been lucky! 2011-05-05T09:51:35 did you see the merged radar map that showed the length and intensity of the storm 2011-05-05T09:51:47 as a student, I also think I should contact with mentor and establish a good relationship through this period :) 2011-05-05T09:51:48 actually not yet 2011-05-05T09:51:58 it is pretty cool and very scary. 2011-05-05T09:52:50 it was strange to find out when we got to Chattanooga for the weekend to stay in a hotel with power that each area had so much damage, we were inwardly focused and had no idea about how bad other places were 2011-05-05T09:52:59 Ralf said it made the paper in Ulm for days 2011-05-05T09:57:53 hmm.. supercell went pretty straight 2011-05-05T09:58:25 That's a LONG path and wide!! pretty cool graphic 2011-05-05T09:58:40 we saw reports of business signs traveling over 100 miles (160km) 2011-05-05T09:59:17 how often do that big tornade happen there? 2011-05-05T09:59:17 Thurs-Sun was nice here 70F.. now 50F.. sadly that type of up and down temps can lead to those storms 2011-05-05T10:00:07 lets.. see.. F to C converter.. 2011-05-05T10:00:18 a few a season but not like that. One neighborhood just outside Huntsville was hit by an F-4 in 1995-ish, worse this time, and someone who grew up here said it got one in 1974 but was a field then 2011-05-05T10:00:26 21C -> 10 C 2011-05-05T10:00:28 sorry.. a bit slow on the conversions today.. 2011-05-05T10:00:46 and this is May here.. should be time to wear shorts :D 2011-05-05T10:00:48 70F is nice (21C) 2011-05-05T10:01:20 50F is .. not :D .. we have that much in here 2011-05-05T10:02:11 we should have 70F here after sunday 2011-05-05T10:03:31 so tornado there is common natural event.. :) 2011-05-05T10:03:52 i have never seen one 2011-05-05T10:05:46 unfortunately not as uncommon as one would like. Normally very localized damage. 2011-05-05T10:06:43 The big issue here was that on top of the "normal" damage, our regional utility lost all high voltage power transmission lines. This is what put us down for a week. North Alabama and Mississippi were impacted. At one point 1M people were without power. 2011-05-05T10:08:05 yep.. everyone should have their own windmill or solarpanel or something:D 2011-05-05T10:08:24 maybe storm proof 2011-05-05T10:09:28 I am on the same power substation as the main hospital here. It got power with my house did -- overnight Sunday. Research Park was put off until Tuesday night. 2011-05-05T10:09:55 NASA Marshall and Redstone Arsenal still have NOT opened. Not all power is back to them. 2011-05-05T10:10:03 :l 2011-05-05T10:10:14 good for you:) 2011-05-05T10:10:24 and hospital :p 2011-05-05T10:10:25 like a Katrina scale power outage with much less destruction. Strange 2011-05-05T10:10:40 selective storm 2011-05-05T10:10:57 yep.. but it selected a lot :-D 2011-05-05T10:11:14 greedy thing 2011-05-05T10:11:24 I am promising this will have a positive impact on the RTEMS Project. 2011-05-05T10:11:43 great attitude 2011-05-05T10:12:17 RTEMS will be more storm proof or self sustained..:D? 2011-05-05T10:19:39 *** xiangfu has quit IRC 2011-05-05T10:29:19 looking into possibilities of another more complete mirror. This storm knocked 2 of the 3 yum/ftp mirrors 2011-05-05T10:29:48 also want to make sure communications paths are remembered. I forgot about IRC topic, linkedin and facebook as ways to get news to community 2011-05-05T10:30:20 We can't prevent disaster but we can broaden our back up plans 2011-05-05T10:31:55 :) 2011-05-05T10:59:15 heh, I was in northern Ohio back in 2003, where much of the northeast lost power due to a single stupid fault - losing power is Not Cool 2011-05-05T10:59:39 I hope that everything is recovering well, and that the last week and some hasn't been too terrible for anyone 2011-05-05T11:05:58 *** sqm has quit IRC 2011-05-05T11:07:10 /8 2011-05-05T11:07:21 *** lj8175 has quit IRC 2011-05-05T11:11:26 DrJoel: what other build machines are free for me to use? i've setup everything on rtbf32a at the moment.. 2011-05-05T11:22:59 hmm meeting, bbl 2011-05-05T11:41:25 rtbf64a and rtbf64b 2011-05-05T12:04:24 *** zwj has quit IRC 2011-05-05T12:33:09 *** DrJoel has quit IRC 2011-05-05T12:47:54 *** DrJoel has joined #rtems 2011-05-05T12:47:54 *** ChanServ sets mode: +o DrJoel 2011-05-05T14:25:10 *** dr__house has joined #rtems 2011-05-05T14:25:10 *** dr__house has joined #rtems 2011-05-05T15:48:42 *** DrJoel changes topic to "RTEMS is proud to have 8 students sponsored by Google in this year's Summer of Code" 2011-05-05T15:55:21 hi 2011-05-05T15:55:28 is there a way to read the IP routing table from the application, except hack into the rt_tables array? 2011-05-05T15:56:26 (which is in net/route.h with #ifdef _KERNEL) 2011-05-05T15:57:41 good question for the mailing list. Probably a good place to provide an API. 2011-05-05T15:58:20 lekernel, I think so. Check the BSD route command. 2011-05-05T15:59:19 yeah, it merely calls rtems_bsdnet_show_inet_routes() in the kernel which in turn uses the rt_tables array *g* 2011-05-05T15:59:47 I mean in the FeeBSD sources. The route command is the one on BSD boxes that does this. 2011-05-05T16:00:04 Another way is the sysctl interface that SNMP uses. 2011-05-05T16:02:57 ah... yeah... seems the rtioctl() is there 2011-05-05T16:03:35 but why is it that the RTEMS shell route command doesn't use it and makes a rather ugly hack instead? 2011-05-05T16:12:26 uhm, oh, well 2011-05-05T16:12:41 in fact it seems the BSD route reads kernel memory directly 2011-05-05T16:13:03 http://koders.com/c/fidE6A14168FAC673EF95A12C9471BBCFC8FDAF6351.aspx?s=route.c#L1 with that kread() call 2011-05-05T16:14:07 The route command was a simple way to test the stack when making it go. 2011-05-05T16:14:07 so... both are equally dirty hacks 2011-05-05T16:14:21 Is that code in the kernel or user land ? 2011-05-05T16:14:37 userland apparently :-) 2011-05-05T16:14:47 Hmmm usr/bin/netstat 2011-05-05T16:14:51 yeah lol 2011-05-05T16:14:56 I didn't think BSD sucked so much 2011-05-05T16:15:10 It must get a copy 2011-05-05T16:15:45 Following the route command, what path it takes will protect your application 2011-05-05T16:15:56 This is the key point. 2011-05-05T16:16:22 This is open BSD. What does FreeBSD do ? 2011-05-05T16:16:39 #define _KERNEL #include #undef _KERNEL 2011-05-05T16:16:42 hum .... 2011-05-05T16:16:52 amazing 2011-05-05T16:17:09 Pays not to look to hard under the hood :) 2011-05-05T16:29:22 *** dr__house has quit IRC 2011-05-05T16:44:21 kiwichris: i don't have an account on rtbf64(a,b) do i? 2011-05-05T16:45:51 verm__: checking 2011-05-05T16:45:57 thank you! 2011-05-05T16:46:10 not on 64a 2011-05-05T16:46:13 it's been so long i forget i had to ask jeff to get me onto rtbf32a 2011-05-05T16:46:34 it has been a very long couple of weeks.. we barely remember where we are 2011-05-05T16:47:40 yeah i hear that 2011-05-05T17:24:15 kiwichris: that reminds me, i forgot to cover one point last night 2011-05-05T17:24:24 Ok 2011-05-05T17:24:44 since the BSPs are a single unit now, i plan to schedule them independently, this means we can trigger only those BSPs that have had their related files changed 2011-05-05T17:24:54 for instance if someone edits a file for a specific BSP *only* that bsp will be re-built 2011-05-05T17:25:11 some files will trigger all bsps to be rebuilt, or perhaps only a certain target 2011-05-05T17:25:20 it's pretty neat, what is not neat is setting it up 2011-05-05T17:25:38 we need to collect regular expressions to divide files into global, target and per-bsp 2011-05-05T17:26:07 it is a simple list of regexps that are checked over the incoming file list (which has full path info) 2011-05-05T17:26:22 This is complicated. For example the testsuite would need the list of valid tests for a BSP. 2011-05-05T17:26:43 i was planning on doing that anyway 2011-05-05T17:27:00 DrJoel, this reminds me we need to sort out the BSP test config in the BSP config dir 2011-05-05T17:27:04 we can rebuild if any test files are changed for now, how about the source files excluding that? 2011-05-05T17:27:06 verm__, great 2011-05-05T17:27:28 is there any simple way to get a general list that will do most of the work? we can fine tune it later 2011-05-05T17:27:56 Starting with cpukit the libnetworking is conditional on --enable-networking which I think is the default. 2011-05-05T17:28:57 ftpd is the same 2011-05-05T17:29:03 that is cpukit/ftpd 2011-05-05T17:29:19 cpukit/include is too hard at this point in time 2011-05-05T17:29:31 cpukit/libblock I suppose all 2011-05-05T17:29:40 cpukit/libcsupport same 2011-05-05T17:29:52 cpukit/libgant I will flick to Joel 2011-05-05T17:29:59 cpukit/libgnat I will flick to Joel 2011-05-05T17:30:11 cpukit/libfs all 2011-05-05T17:30:18 cpukit/libi2c all 2011-05-05T17:30:29 cpukit/libmd all 2011-05-05T17:30:36 cpukit/libmisc all 2011-05-05T17:30:37 kiwichris: yes on test config.. whatever that turns into.. as well as my work on getting the script for exe->ralf worked out 2011-05-05T17:30:54 libgnat.. ignore unless building Ada. So ignore for now 2011-05-05T17:31:02 cpukit/librpc is under networking 2011-05-05T17:31:18 let's reversse your logic chris. 2011-05-05T17:31:20 cpukit/libmghttpd is under networking 2011-05-05T17:31:53 Ok then we have networking and exclude 2011-05-05T17:32:22 anything under score/cpu/ARCH, c/src/lib/libcpu/ARCH, or libbsp/ARCH/shared is architecture dependent and should trigger all BSPs in that architecture 2011-05-05T17:32:23 Plus also under cpukit/score/cpu we have the target 2011-05-05T17:33:07 libbsp/shared should probably trigger all BSPs 2011-05-05T17:33:24 Yeap 2011-05-05T17:33:53 Virtually anything else should trigger all.. except for feature flag issues like networking 2011-05-05T17:34:30 Therefore is posix excluded on --disable-posix ? 2011-05-05T17:35:12 yes (except for a few files). I think mghttpd required posix and networking to be enabled 2011-05-05T17:36:02 hmm 2011-05-05T17:36:12 You did ask :) 2011-05-05T17:36:57 autoconf had host, target, and multilib... 2011-05-05T17:37:11 we add specific CPU model and board to that mix.. 2011-05-05T17:37:15 what about non-source files like doc files and whatnot? 2011-05-05T17:37:32 under testsuites .. no impact 2011-05-05T17:37:37 and how do i target per-bsp files or per-target files? 2011-05-05T17:37:46 under /docs at the top need to build just docs 2011-05-05T17:38:06 o 2011-05-05T17:38:12 i'll make a specific builder just for docs then 2011-05-05T17:38:23 you mean how do you select arch and BSP? 2011-05-05T17:38:40 --target=CPU-rtemsVERSION like sparc-rtems4.11 2011-05-05T17:38:42 i've already got that sorted 2011-05-05T17:38:52 board --enable-rtemsbsp=sis 2011-05-05T17:38:58 if someone modifies a file that only affects "arm" 2011-05-05T17:39:09 for example is there an arm/common 2011-05-05T17:39:16 or arm/arm1136jfs/ 2011-05-05T17:39:17 ? 2011-05-05T17:39:17 say in score/cpu/arm or c/src/lib/libcpu/arm 2011-05-05T17:39:33 the first would trigger all arm bsps to be built 2011-05-05T17:39:42 while the second only the arm1136jfs bsp 2011-05-05T17:39:55 there's no sense in wasting cputime building things that were not touched in a commit... 2011-05-05T17:40:13 agree 100% .. 2011-05-05T17:40:40 this also gives us loose parallelism 2011-05-05T17:40:43 the actual requirements for a BSP need to know the CPU model.. that is in libbsp/CPU/*/make/custom/BSP.cfg (maybe with an include to process) 2011-05-05T17:41:06 for instance if you commit to arm and chris commits to sparc both can build at the same time.. the 'arm' commit won't make the sparc builders busy since it'll ignore any arm commits 2011-05-05T17:41:22 and then what that CPU model implies will be built in c/src/lib/libcpu/CPU.. that logic is in c/src/lib/libcpu/*/configure.ac 2011-05-05T17:43:13 hmm i see 2011-05-05T17:43:29 so the only per-target / per-bsp files live in c/src/lib/libcpu? 2011-05-05T17:43:59 also the .cfg files in c/src/lib/libbsp 2011-05-05T17:44:19 not quite.. libcpu are specific to a CPU model.. say UART on system on chip (SOC) or DMA or cache management 2011-05-05T17:44:39 libbsp/CPU/BSP.... is supposed to have the minimum set of files specific to that single board 2011-05-05T17:45:04 BUT... since there are often incompatible variants of boards, there can be BSP "variants" 2011-05-05T17:45:23 so .... architecture/target .. then cpu model.. then BSP family.. then BSP 2011-05-05T17:46:29 you can subdivide far but I don't know if the payoff is actually worth it given the way changes tend to happen. 2011-05-05T17:48:01 kiwichris: would you agree that much past (1) score/cpu, c/scr/lib/libcpu, or libbsp/CPU/shared triggering all BSPs within arch and (2) libbsp/CPU/BSP_FAMILY triggering all BSPs within the family, there is likely not much payoff? 2011-05-05T17:48:28 The source that tends to change is architecture independent 2011-05-05T17:49:06 Yes. The issue is time to build verses the time to make a fine grain break down. Is it worth it. 2011-05-05T17:52:13 I think the only accurate way to do it would be to parse the make/custom files and libcpu/*/configure.ac files 2011-05-05T17:53:35 I am of the "simple and working" first. Is there a log of what triggered the builds? We could start with something like I described and then watch it for a while and see if it is really triggering too much most of the time. 2011-05-05T17:55:02 * kiwichris will be back soon 2011-05-05T17:59:07 DrJoel: sure i'm fine with doing something more general and narrowing it down over time 2011-05-05T17:59:25 buildbot will show the list of files in the change and it's easy to see what was triggered 2011-05-05T18:00:28 you're right that getting too fine grained can be pointless, it'll depend on how long building ends up being on the rtbf32a and rtbf64(a,b) 2011-05-05T18:01:21 Our change rate is typically not that high and those are reasonably fast machines 2011-05-05T18:05:18 ok 2011-05-05T18:05:30 btw where is the master repository that gets the initial commits? 2011-05-05T18:17:01 /usr1/CVS on rtems.org 2011-05-05T18:19:41 oh so that is the master? ok 2011-05-05T18:19:59 i need to figure out how to get the changes to the master which is on rtbf32a 2011-05-05T18:20:16 there's a hook script that needs to be installed in the repo, which sends an email 2011-05-05T19:11:48 *** DrJoel has quit IRC 2011-05-05T20:25:55 *** xiangfu has joined #rtems 2011-05-05T20:32:43 *** xiangfu has quit IRC 2011-05-05T20:51:07 *** kristianpaul has quit IRC 2011-05-05T20:54:46 *** kristianpaul has joined #rtems 2011-05-05T20:54:46 *** kristianpaul has joined #rtems 2011-05-05T21:07:45 *** xiangfu has joined #rtems 2011-05-05T21:15:49 *** sqm has joined #rtems 2011-05-05T22:24:58 *** xiangfu has quit IRC 2011-05-05T23:54:21 *** xiangfu has joined #rtems 2011-05-06T00:22:37 *** monstr has joined #rtems 2011-05-06T00:29:36 *** kiwichris has quit IRC 2011-05-06T01:00:55 *** sebhub has joined #rtems 2011-05-06T01:01:03 good morning 2011-05-06T01:04:06 hi 2011-05-06T01:46:09 *** peerst has left #rtems 2011-05-06T01:48:55 *** sqm has quit IRC 2011-05-06T02:34:28 *** monstr has quit IRC 2011-05-06T03:07:58 *** arvind_khadri has joined #rtems 2011-05-06T04:42:40 *** xiangfu has quit IRC 2011-05-06T05:32:44 *** sqm has joined #rtems 2011-05-06T06:41:29 *** sebhub has quit IRC 2011-05-06T06:59:50 *** lekernel has quit IRC 2011-05-06T06:59:52 *** lekernel_ has joined #rtems 2011-05-06T07:23:12 *** sebhub has joined #rtems 2011-05-06T07:25:22 excerpt from the RTEMS DHCP client: http://pastebin.com/b8qa6at1 2011-05-06T07:25:28 rtems_bsdnet_ntpserver ??? 2011-05-06T07:25:33 (line 3) 2011-05-06T07:26:47 also, shouldn't res_init() be called for the new nameservers to be taken into account? 2011-05-06T07:27:17 when the DHCP client is started _after_ rtems_bsdnet_initialize_network() 2011-05-06T07:29:26 but ... it doesn't seem to be that res_init() implements any form of locking 2011-05-06T07:29:27 argh 2011-05-06T07:36:54 *** lekernel has joined #rtems 2011-05-06T07:39:47 *** lekernel_ has quit IRC 2011-05-06T08:01:49 *** lj8175 has joined #rtems 2011-05-06T08:35:02 *** xiangfu has joined #rtems 2011-05-06T08:59:23 *** lj8175 has quit IRC 2011-05-06T09:35:02 *** lj8175 has joined #rtems 2011-05-06T09:41:04 *** JenniferAverett has joined #rtems 2011-05-06T09:44:11 *** arvind_khadri has quit IRC 2011-05-06T09:55:51 *** gedare has joined #rtems 2011-05-06T10:03:09 *** lj8175 has quit IRC 2011-05-06T10:08:15 *** lj8175 has joined #rtems 2011-05-06T10:18:30 *** xiangfu has quit IRC 2011-05-06T11:21:31 *** lj8175 has quit IRC 2011-05-06T11:26:46 *** sebhub has quit IRC 2011-05-06T11:54:42 *** gedare has quit IRC 2011-05-06T12:06:27 *** sqm has quit IRC 2011-05-06T12:42:44 *** arvind_khadri has joined #rtems 2011-05-06T13:23:26 *** arvind_k has joined #rtems 2011-05-06T13:23:40 *** arvind_khadri has quit IRC 2011-05-06T14:02:47 *** arvind_k has quit IRC 2011-05-06T14:40:41 *** JenniferAverett has quit IRC 2011-05-06T16:18:46 *** kiwichris has joined #rtems 2011-05-06T18:23:15 *** kristianpaul has quit IRC 2011-05-06T18:23:37 *** kristianpaul has joined #rtems 2011-05-06T19:19:51 *** easwar has joined #rtems 2011-05-06T19:19:51 *** easwar has joined #rtems 2011-05-06T19:31:32 kiwichris: rtbf64a doesn't have m32r-rtems4.11-gcc installed can you fix this? 2011-05-06T19:31:49 Sure. Hang on. 2011-05-06T19:33:54 thanks 2011-05-06T19:34:43 Just 4.11 2011-05-06T19:34:51 ? 2011-05-06T19:35:05 yep 2011-05-06T19:38:11 Done 2011-05-06T19:39:54 thanks 2011-05-06T19:40:08 I saw the failure on that BSP. 2011-05-06T19:40:11 kiwichris: do you have tiem this evening to help me out with some things? 2011-05-06T19:40:34 It is Sat morning here so at the moment but that may change :) 2011-05-06T19:41:01 oh right i forgot it's friday 2011-05-06T19:41:27 did you hear about the new build box? 2011-05-06T19:41:40 Yeah I did see something. Nice 2011-05-06T19:41:59 I looked over the buildbot results. Nice. 2011-05-06T19:42:43 cool, i ran into an annoying bug in buildbot which soaked up several hours 2011-05-06T19:42:54 Ouch 2011-05-06T20:23:04 kiwichris: we should setup a mailing list for buildbot failures 2011-05-06T20:23:11 there's also an irc bot... 2011-05-06T20:36:00 Yes and I think yes. 2011-05-06T20:36:21 Ping Jeff for the mailing list. What is needed for the IRC bot ? 2011-05-06T20:36:38 nothing i can launch the irc bot directly from the machine 2011-05-06T20:36:44 it runs directly from the master 2011-05-06T20:36:50 as a thread 2011-05-06T20:37:04 easy. 2011-05-06T20:38:46 *** rtemsbuild has joined #rtems 2011-05-06T20:38:55 that easy 2011-05-06T20:39:12 What can it do ? 2011-05-06T20:39:21 rtemsbuild: force build sparc sis VANILLA 2011-05-06T20:39:21 no such builder 'sparc' 2011-05-06T20:39:23 hmm 2011-05-06T20:39:43 i wonder if it's having issues with the spaces it has some issues since noone really works on it 2011-05-06T20:39:44 Oh I see. That is nice. 2011-05-06T20:39:51 rtemsbuild: force build "sparc sis VANILLA" 2011-05-06T20:39:56 The build has been queued, I'll give a shout when it starts 2011-05-06T20:39:59 build #1 of sparc sis VANILLA is complete: Exception [exception download] Build details are at http://buildbot.rtems.org/builders/sparc%20sis%20VANILLA/builds/1 2011-05-06T20:40:19 Cool. Can it report the status of a build ? 2011-05-06T20:40:23 yep 2011-05-06T20:40:37 I see that as powerful 2011-05-06T20:40:49 yeah there's a lot of cool features 2011-05-06T20:40:56 I am off out for a while. 2011-05-06T20:40:58 also that exception is another bug, it has to do with us running cvs 2011-05-06T20:41:01 it's been a real pain 2011-05-06T20:41:07 I will back later today. 2011-05-06T20:41:12 yep have fun 2011-05-06T20:41:15 I understand. 2011-05-06T20:41:34 I have put you forward to do the conversion :) 2011-05-06T20:41:49 heh i'm fine with it since it'll make my life easier 2011-05-06T20:41:57 i'd rather spend it trying to convince people to migrate than fighting cvs :) 2011-05-06T20:42:15 *** rtemsbuild has quit IRC 2011-05-06T20:42:24 *** rtemsbuild has joined #rtems 2011-05-06T20:42:28 rtemsbuild: force build "sparc sis VANILLA" 2011-05-06T20:42:29 build #2 of sparc sis VANILLA is complete: Exception [exception download] Build details are at http://buildbot.rtems.org/builders/sparc%20sis%20VANILLA/builds/2 2011-05-06T20:42:29 Agreed and it is a valid reason why we need to move. 2011-05-06T20:42:34 The build has been queued, I'll give a shout when it starts 2011-05-06T20:42:34 I am off. 2011-05-06T20:42:39 yep 2011-05-06T20:57:49 *** xiangfu has joined #rtems 2011-05-06T22:46:54 *** sqm has joined #rtems 2011-05-06T23:00:21 *** easwar has left #rtems 2011-05-06T23:07:06 *** xiangfu has quit IRC 2011-05-06T23:15:41 *** xiangfu has joined #rtems 2011-05-06T23:25:48 *** xiangfu has quit IRC 2011-05-07T00:03:39 *** sqm has quit IRC 2011-05-07T00:22:01 *** sqm has joined #rtems 2011-05-07T00:42:38 *** sqm has quit IRC 2011-05-07T02:50:54 *** rtemsbuild has quit IRC 2011-05-07T03:08:37 *** xiangfu has joined #rtems 2011-05-07T05:12:09 *** xiangfu has quit IRC 2011-05-07T05:25:40 *** xiangfu has joined #rtems 2011-05-07T05:38:21 kiwichris: let me know when you're around.. 2011-05-07T07:14:20 *** sqm has joined #rtems 2011-05-07T07:27:57 here for a moment 2011-05-07T07:30:03 hmm well, i need some starting points for that change list i talked about the other day 2011-05-07T07:30:21 so we can avoid building certain bsps when the change is unrelated 2011-05-07T07:36:16 *** arvind_khadri has joined #rtems 2011-05-07T07:36:43 Ok. It is late here and a long evening but I understand. I think Joel outlines something that I need to read. 2011-05-07T07:38:21 no worries it's the weekend, i'm just trying to do this while i have free time 2011-05-07T07:38:28 there's other things i can work on 2011-05-07T07:38:52 *** lekernel has quit IRC 2011-05-07T07:43:19 I understand. I think the CVS issue it critical. 2011-05-07T07:43:58 How clean is the repo from CVS you have ? 2011-05-07T07:46:55 *** lekernel has joined #rtems 2011-05-07T07:48:22 the git one i made? 2011-05-07T07:48:33 the comments about it are in the email i sent 2011-05-07T07:48:43 i need to re-add the branches 2011-05-07T07:48:58 and fix the committer names both of which are trivial to fix 2011-05-07T07:49:20 kiwichris: i can send you a list of names i need if you give those to me i can fix them 2011-05-07T07:50:13 I am answering the email. I feel the ability to capture RTEMS history in a conversion will be a critical issue. 2011-05-07T07:51:02 I think so. The SC contains the history to it can happen. 2011-05-07T07:51:04 all the history is there 2011-05-07T07:51:13 All branches ? 2011-05-07T07:51:16 yes 2011-05-07T07:51:21 Fantastic. 2011-05-07T07:51:27 'git branch' will not show anything use 'git tag' 2011-05-07T07:51:40 Ralf is away at the moment. I will not move until he has said ok. 2011-05-07T07:51:42 we can re-create the version branches as we need them 2011-05-07T07:52:36 Is the repo the one you sent in the email the other day ? 2011-05-07T07:52:45 That the buildbot uses ? 2011-05-07T07:54:34 yeah i'm making the changes to buildbot now to use it 2011-05-07T07:54:42 it is in my homedir on the rtems machine ~/rtems.git 2011-05-07T07:54:50 do not use the one on rtbf32a 2011-05-07T07:55:03 Yeap. I will let Ralf know and ask him to have a look. 2011-05-07T07:55:17 in that case i should fix it up so it is OK for development 2011-05-07T07:55:30 I am off for the night. Have a good weekend. 2011-05-07T08:01:35 ok, i sent some emails so i can finish fixing the repo 2011-05-07T08:08:41 *** sqm has quit IRC 2011-05-07T08:11:09 *** arvind_khadri has quit IRC 2011-05-07T08:33:34 *** xiangfu has quit IRC 2011-05-07T09:13:27 *** peerst has joined #rtems 2011-05-07T09:13:39 *** peerst has left #rtems 2011-05-07T09:14:09 *** peerst has joined #rtems 2011-05-07T09:33:11 *** arvind_khadri has joined #rtems 2011-05-07T09:42:19 *** dimonov has quit IRC 2011-05-07T09:43:22 *** dimonov has joined #rtems 2011-05-07T09:57:29 *** dimonov has quit IRC 2011-05-07T09:58:37 *** dimonov has joined #rtems 2011-05-07T10:08:47 *** arvind_khadri has quit IRC 2011-05-07T10:18:22 *** arvind_khadri has joined #rtems 2011-05-07T11:18:20 *** arvind_khadri has quit IRC 2011-05-07T11:18:54 *** arvind_khadri has joined #rtems 2011-05-07T11:29:36 *** arvind_khadri has quit IRC 2011-05-07T11:42:34 *** arvind_khadri has joined #rtems 2011-05-07T11:54:43 *** sqm has joined #rtems 2011-05-07T14:53:25 *** arvind_khadri has quit IRC 2011-05-07T15:03:41 *** shayden has quit IRC 2011-05-07T15:12:03 *** dr__house has joined #rtems 2011-05-07T15:12:03 *** dr__house has joined #rtems 2011-05-07T15:49:57 *** dimonov has quit IRC 2011-05-07T15:50:49 *** dimonov has joined #rtems 2011-05-07T15:50:49 *** dimonov has joined #rtems 2011-05-08T00:17:40 *** arvind_khadri has joined #rtems 2011-05-08T00:25:19 *** arvind_khadri has quit IRC 2011-05-08T00:32:49 *** xiangfu has joined #rtems 2011-05-08T07:40:50 *** lekernel has quit IRC 2011-05-08T07:53:07 *** lekernel has joined #rtems 2011-05-08T08:16:34 *** lj8175 has joined #rtems 2011-05-08T10:41:40 *** xiangfu has quit IRC 2011-05-08T10:58:16 *** lj8175 has quit IRC 2011-05-08T11:15:10 *** lekernel has quit IRC 2011-05-08T18:05:54 *** projectgus has joined #rtems 2011-05-08T19:06:13 verm__, hi 2011-05-08T19:06:28 hey 2011-05-08T19:06:45 Can you walk me through the tag/branch topic ? 2011-05-08T19:07:06 i responded to your email, did you see it? 2011-05-08T19:07:17 Yes and I still need help :) 2011-05-08T19:07:22 hehe ok 2011-05-08T19:07:49 a different way to explain it is if we have a 4.11 branch 2011-05-08T19:07:57 We publicly maintain the head (master) and the previous 2 releases. 2011-05-08T19:08:08 and make commits to it, after we do a release we tag 4.11 to that specific version 2011-05-08T19:08:10 There are changes and releases made on these. 2011-05-08T19:08:19 if we are never making changes to a version it does not need to be a branch 2011-05-08T19:08:41 kiwichris: right so we'll need branches for 4.10 and 4.9 while 4.11 is 'master' correct? 2011-05-08T19:08:47 Yes 2011-05-08T19:09:06 OAR also provides a service for other releases 2011-05-08T19:09:25 They do get requests for 10 year old releases. 2011-05-08T19:09:25 you can trivially add/remove branches as you want with git 2011-05-08T19:09:47 Ok. 2011-05-08T19:09:50 there's no need to keep them around just make a tag of it and delete the branch you can alwys re-create it with a single command later 2011-05-08T19:10:09 So for 4.6 I would recreate the branch do some work commit the changes 2011-05-08T19:10:14 it's not like CVS or SVN where it a long operation 2011-05-08T19:10:37 kiwichris: yeah, you can do two things, create the branch from the most recent 4.6 tag 2011-05-08T19:10:44 or create it from a specific revision 2011-05-08T19:11:05 What does "specific revision" mean ? 2011-05-08T19:11:49 exaclty what it sounds like, choose any revision in the tree 2011-05-08T19:12:00 In git what is this ? 2011-05-08T19:12:03 and you can make a branch right off of it, remember changes are atomic it's not like CVS where it's per-file 2011-05-08T19:12:22 Yes which is the most important feature needed in the change. 2011-05-08T19:12:25 those long hashes 2011-05-08T19:12:39 Ah ok a machine all knowing label. 2011-05-08T19:12:52 901dac646d9f6c51e5bb1112a72f50d1f63fd697 for example 2011-05-08T19:13:13 I know understand. 2011-05-08T19:13:47 So we need development work flow, maintained releases work flow and archived releases work flow. 2011-05-08T19:14:10 it'll be much easier with git. 2011-05-08T19:14:20 I am thinking about dooc 2011-05-08T19:14:57 What VC, a documented work flow puts it in the lead. 2011-05-08T19:15:03 What even VC ... 2011-05-08T19:15:11 What ever VC ... (grr) 2011-05-08T19:15:59 Ralf is away this weekend. He will back tomorrow I think 2011-05-08T19:16:55 well, this is one of the things i get paid to do, helping out in this area is not a problem 2011-05-08T19:17:10 I broke 4.10 and 4.11 with the PR 1774 change. The buildbit with SIS testing would have would it 2011-05-08T19:17:16 i rarely run into a case where i do not have to change a clients RCS in order to install a proper testing system 2011-05-08T19:17:36 usually those who have good testing already use a sane RCS as they have run into the problems i usually run into... 2011-05-08T19:18:08 PR 1774 is a nice example of needing the feedback rather than someone coming across the regression and sending me an email 2011-05-08T19:18:16 kiwichris: yeah, and with the branch building i can have buildbot watch a specific branch 2011-05-08T19:18:26 Which is very nice. 2011-05-08T19:18:35 for instance if you commit to a test_XXX branch buildbot will do a run on it 2011-05-08T19:18:43 if it passes you can then merge test_XXX into master 2011-05-08T19:18:53 We only need head plus 2 releases, or is it one release. I cannot rememer 2011-05-08T19:20:12 I think tags can be cleaned up. I will let Joel comment on this. 2011-05-08T19:22:12 did you see the full list? 2011-05-08T19:22:22 No I have not. 2011-05-08T19:22:40 git --git-dir=/home/amar/rtems.git tag 2011-05-08T19:22:45 on rtems will show it 2011-05-08T19:22:56 there's a lot.. 2011-05-08T19:24:11 Says it is not a repository ! 2011-05-08T19:26:05 Is that on rtems.org ? 2011-05-08T19:26:09 yeah rtems.org 2011-05-08T19:26:23 i wonder if the read permissions are wrong, hmmm 2011-05-08T19:26:26 where were you running it? 2011-05-08T19:26:33 Yeap 2011-05-08T19:26:56 Ok got. 2011-05-08T19:27:38 I suspect the MOVED* and rtems-ss-* can go 2011-05-08T19:27:44 ss = snapshot 2011-05-08T20:09:40 *** xiangfu has joined #rtems 2011-05-08T21:22:01 kiwichris: ah 2011-05-08T21:22:04 cool 2011-05-08T23:48:33 *** sqm has joined #rtems